KR102421720B1 - Apparatus and method for performing artificial intelligence encoding and artificial intelligence decoding of image - Google Patents

Abstract

a video bitstream including at least one processor executing one or more instructions, wherein the processor includes a main bitstream including image data obtained as a result of the first encoding of a first image and a sub bitstream including AI data obtain a second image corresponding to the first image by first decoding image data included in the main bitstream, acquire AI upscaling activation flag information included in AI data of the sub-bitstream, and AI It is determined whether to perform AI upscaling on the second image based on the upscaling activation flag information, and when it is determined to perform AI upscaling on the second image, at least a portion of image data and AI included in the sub-bitstream Obtaining an AI upscaled third image from the second image through the upscaling DNN set according to the selected upscaling DNN information among a plurality of upscaling DNN information stored in advance based on at least one of at least one of the sub data and, when the third image is obtained, an AI decoding apparatus according to an embodiment for outputting the third image is disclosed.

Description

Apparatus and method for APA encoding and AIA decoding of an image TECHNICAL FIELD

The present disclosure relates to the field of encoding and decoding of images. More specifically, the present disclosure relates to an apparatus and method for encoding/decoding an image based on AI.

The image is encoded by a codec conforming to a predetermined data compression standard, for example, a Moving Picture Expert Group (MPEG) standard, and then stored in a bitstream in the form of a recording medium or transmitted through a communication channel.

With the development and dissemination of hardware capable of reproducing and storing high-resolution/high-definition images, the need for a codec capable of effectively encoding and decoding high-resolution/high-definition images is increasing.

An apparatus and method for AI encoding and AI decoding of an image according to an embodiment have a technical task of encoding and decoding an image based on AI to achieve a low bit rate.

AI decoding apparatus according to an embodiment, comprising at least one processor executing one or more instructions,

The processor is

Obtaining a video bitstream including a main bitstream including image data obtained as a result of the first encoding of the first image and a sub-bitstream including AI data,

obtaining a second image corresponding to the first image by first decoding image data included in the main bitstream;

obtaining AI upscaling activation flag information included in the AI data of the sub-bitstream, and determining whether to perform AI upscaling on the second image based on the AI upscaling activation flag information;

When it is determined to perform AI upscaling on the second image,

Upscaling DNN set according to selected upscaling DNN information from among a plurality of upscaling DNN information stored in advance based on at least one of at least one of at least a part of the image data and at least one of the AI sub data included in the sub bitstream Through, to obtain an AI upscaled third image from the second image,

When the third image is obtained, output the third image,

When it is determined not to perform the AI upscaling on the second image, the second image is output,

The AI upscaling activation flag information may indicate whether AI upscaling is performed on the second image.

The AI sub data included in the sub-bitstream includes image genre information indicating the genre of the image, average quantization parameter information indicating the average quantization value of the first image, and quantization indicating the quantization degree of at least a part of the first image. parameter indicator information, image resolution information indicating the resolution of at least one of the original image and the first image, image ratio information indicating the ratio of the original image and the first image, and the codec used for encoding the first image It may include at least one of codec information indicating, metadata flag information indicating whether to include additional metadata related to AI upscaling, and additional metadata information related to AI upscaling.

The processor is

Among a plurality of upscaling DNN information based on at least one of the image genre information, the average quantization parameter information, the quantization parameter indicator information, the image resolution information, the image ratio information, and the AI upscaling related additional metadata information One piece of DNN for upscaling may set the upscaling DNN.

The quantization parameter indicator information may be index information indicating one quantization parameter group among a plurality of predetermined quantization parameter groups,

The processor may set the upscaling DNN as upscaling DNN information corresponding to one quantization parameter group indicated by the index information among a plurality of upscaling DNN information.

The quantization parameter indicator information may be index information indicating a quantization parameter group including an (average) value of at least one quantization parameter among a subgroup in a frame, a frame, a frame group, and a video sequence.

The processor may set the upscaling DNN as selected upscaling DNN information from among a plurality of upscaling DNN information based on quantization parameter information obtained from the image data of the main bitstream.

The main bitstream may be included in a predetermined main bitstream area in a video bitstream defined by a predetermined codec, and the sub bitstream may be included in a predetermined sub bitstream area in a video bitstream defined in a predetermined codec. .

The sub-bitstream may be a Supplemental Enhancement Information (SEI) message.

The SEI message may be an SEI message including user data registered according to a predetermined standard or unregistered user data identified by a Universally Unique Identifier (UUID).

The sub-bitstream may be metadata Open Bitstream Unit (OBU).

AI encoding apparatus according to an embodiment,

at least one processor executing one or more instructions;

The processor is

Decide whether to downscale the original video by AI,

Based on the determination result, an AI downscaled first image is obtained from the original image using a downscale DNN,

Obtaining image data by first encoding the first image,

generating AI upscaling activation flag information indicating whether AI upscaling corresponding to AI downscaling of the original video;

If it is determined to perform the AI downscaling on the original image, generate AI sub data related to AI upscaling corresponding to the AI downscaling,

generating AI data including the AI upscale activation flag information and the AI sub data;

generating a video bitstream including a main bitstream including the image data and a sub bitstream including the AI data,

transmit the video bitstream;

At least one of at least a part of the image data and at least a part of the AI sub data related to the AI upscaling is information used to select DNN setting information for upscaling from among a plurality of pre-stored DNN setting information for upscaling,

The DNN for upscaling used for AI upscaling may be set as the selected DNN for upscaling configuration information.

AI encoding method according to an embodiment,

determining whether to downscale the original video by AI;

obtaining an AI downscaled first image from the original image using a downscaling DNN based on the determination result;

obtaining image data by first encoding the first image;

Generates AI upscaling activation flag information indicating whether AI upscaling corresponding to AI downscaling of the original video, and when it is determined to perform the AI downscaling on the original video, corresponding to the AI downscaling generating AI sub data related to AI upscaling;

generating AI data including the AI upscale activation flag information and the AI sub data;

generating a video bitstream including a main bitstream including the image data and a sub bitstream including the AI data; and

transmitting the video bitstream;

At least one of at least a part of the image data and at least a part of the AI sub data related to the AI upscaling is information used to select DNN setting information for upscaling from among a plurality of pre-stored DNN setting information for upscaling,

The DNN for upscaling used for AI upscaling may be set as the selected DNN for upscaling configuration information.

AI decoding method according to an embodiment,

obtaining a video bitstream including a main bitstream including image data obtained as a result of a first encoding of a first image and a sub bitstream including AI data;

obtaining a second image corresponding to the first image by first decoding image data included in the main bitstream;

Obtaining AI upscaling activation flag information included in AI data of the sub-bitstream,

determining whether to perform AI upscaling on the second image based on the AI upscaling activation flag information;

When it is determined to perform AI upscaling on the second image,

Upscaling DNN set according to selected upscaling DNN information from among a plurality of upscaling DNN information stored in advance based on at least one of at least one of at least a part of the image data and at least one of the AI sub data included in the sub bitstream through, acquiring an AI upscaled third image from the second image;

outputting the third image when the third image is obtained; and

outputting the second image when it is determined not to perform the AI upscaling on the second image;

The AI upscaling activation flag information may indicate whether AI upscaling is performed on the second image.

AI decoding apparatus according to an embodiment,

at least one processor executing one or more instructions;

The processor is

Obtaining a video bitstream including image data and AI data obtained as a result of the first encoding of the first image,

obtaining a second image corresponding to the first image by first decoding the image data;

obtaining AI upscaling activation flag information included in the AI data, and determining whether to perform AI upscaling on the second image based on the AI upscaling activation flag information;

When it is determined to perform AI upscaling on the second image,

Based on at least one of at least a portion of the image data and at least a portion of the AI sub-data, through the upscaling DNN set according to the upscaling DNN information selected from among a plurality of pre-stored upscaling DNN information, the second image Acquire the AI upscaled third image from

When the third image is obtained, output the third image,

When it is determined not to perform the AI upscaling on the second image, the second image is output,

The AI upscaling activation flag information indicates whether AI upscaling is performed on the second image,

The AI data is included in a header of a predetermined data unit including an encoding parameter related to the image data.

AI decoding apparatus according to an embodiment,

at least one processor executing one or more instructions;

The processor is

Obtaining a video file including AI-encoded data including image data obtained as a result of the first encoding of the first image and AI data,

obtaining a second image corresponding to the first image by first decoding the image data;

obtaining AI upscaling activation flag information included in the AI data, and determining whether to perform AI upscaling on the second image based on the AI upscaling activation flag information;

When it is determined to perform AI upscaling on the second image,

From the second image through the upscaling DNN set according to the selected upscaling DNN information from among a plurality of upscaling DNN information stored in advance based on at least one of at least one of at least a portion of the image data and at least one of the AI sub data Acquire the AI upscaled third image,

When the third image is obtained, output the third image,

When it is determined not to perform the AI upscaling on the second image, the second image is output,

The AI upscaling activation flag information indicates whether AI upscaling is performed on the second image,

The video file may include: a media data box including the image data; and a metadata box including metadata regarding the image data, wherein the metadata box includes the AI data.

AI encoding apparatus according to an embodiment,

at least one processor executing one or more instructions;

The processor determines whether AI downscale of the original image,

Based on the determination result, an AI downscaled first image is obtained from the original image using a downscale DNN,

Obtaining image data by first encoding the first image,

generating AI upscaling activation flag information indicating whether AI upscaling corresponding to AI downscaling of the original video;

When it is determined to perform the AI downscaling on the original image, generate AI sub data related to AI upscaling corresponding to the AI downscaling,

generating AI data including the AI upscale activation flag information and the AI sub data;

Transmitting a video file including AI-encoded data including the image data and the AI data,

At least one of at least a part of the image data and at least a part of the AI sub data related to the AI upscaling is information used to select DNN setting information for upscaling from among a plurality of pre-stored DNN setting information for upscaling,

The DNN for upscaling used for AI upscaling is set as the selected DNN for upscaling setting information,

The video file may include: a media data box including the image data; and a metadata box including metadata regarding the image data, wherein the metadata box includes the AI data.

A computer-recordable recording medium storing a video bitstream including AI-encoded data according to an embodiment, the video bitstream comprising:

a main bitstream including encoding information of a first image obtained by AI downscaling the original image; and a sub-bitstream including AI data for AI upscaling of a second image, wherein the second image is obtained by decoding encoding information of a first image included in the main bitstream. is the corresponding video of

The sub-bitstream is

AI upscaling activation flag information indicating whether AI upscaling is performed on the second image; and

When AI upscaling of the second image is performed, AI sub data used to select upscaling DNN information from among a plurality of pre-stored upscaling DNN information may be included. Computer according to an embodiment The readable recording medium may record a program for executing the disclosed method.

In order to more fully understand the drawings cited herein, a brief description of each drawing is provided.
1 is a diagram for explaining an AI encoding process and an AI decoding process according to an embodiment.
2 is a block diagram illustrating a configuration of an AI decoding apparatus according to an embodiment.
3 is an exemplary diagram illustrating a second DNN for AI upscaling of a second image.
4 is a diagram for explaining a convolution operation using a convolution layer.
5 is an exemplary diagram illustrating a mapping relationship between various pieces of image-related information and various pieces of DNN configuration information.
6 is a diagram illustrating a second image composed of a plurality of frames.
7 is a block diagram illustrating a configuration of an AI encoding apparatus according to an embodiment.
8 is an exemplary diagram illustrating a first DNN for AI downscaling of an original video.
9 is a diagram illustrating a structure of AI encoded data according to an embodiment.
10 is a diagram illustrating a structure of AI encoded data according to another embodiment.
11 is a diagram for explaining a method of training a first DNN and a second DNN.
12 is a diagram for explaining a training process of a first DNN and a second DNN by a training apparatus.
13A is a block diagram illustrating a configuration of an AI decoding apparatus 1300 according to an embodiment.
13B is a block diagram illustrating a configuration of an AI decoding apparatus 1350 according to another embodiment.
14A is a block diagram illustrating a configuration of an AI encoding apparatus 1400 according to an embodiment.
14B is a block diagram illustrating the configuration of an AI encoding apparatus 1450 according to another embodiment.
15A is a flowchart illustrating an AI decoding method according to an embodiment.
15B is a flowchart illustrating an AI decoding method according to another embodiment.
16A is a flowchart illustrating an AI encoding method according to an embodiment.
16B is a flowchart illustrating an AI encoding method according to another embodiment.
17 illustrates the corresponding relationship between the image data 1700 and the AI data 1740, along with the structures of the image data 1700 and the AI data 1740. In FIG.
18 illustrates an embodiment of a video bitstream including AI data and image data.
19 shows an embodiment of AI encoded data when AI data and image data are separated in a single file.
20A illustrates an example of AI encoded data when AI data is inserted into image data.
20B illustrates another embodiment of AI encoded data when AI data is inserted into image data.
20C illustrates an embodiment of AI-encoded data when some AI data is inserted into image data and the remaining AI data is separated from image data.
21 illustrates an embodiment of AI encoded data divided into video segment units when AI data and image data are separated as shown in FIG. 19 .
22 illustrates an embodiment of AI data 2240 and image data 200 transmitted as two separate files.
23 is a diagram illustrating a structure of an SEI message according to an embodiment.
24 illustrates a syntax structure table including AI data, according to an embodiment.
25 is a diagram for explaining operations of a bitstream-based AI encoding apparatus and an AI decoding apparatus according to an embodiment.
26 is an exemplary diagram illustrating a mapping relationship between various pieces of image-related information and various pieces of DNN configuration information.
27 is a diagram for explaining determining indicator (or level) information of a quantization parameter based on a representative QP value of at least a part of an image, according to an embodiment.
28 is an exemplary diagram illustrating a mapping relationship between a QP value and various pieces of DNN configuration information.

Since the present disclosure can make various changes and can have various embodiments, specific embodiments are illustrated in the drawings and will be described in detail through the detailed description. However, this is not intended to limit the embodiments of the present disclosure, and it should be understood that the present disclosure includes all modifications, equivalents and substitutes included in the spirit and scope of various embodiments.

In describing the embodiment, if it is determined that a detailed description of a related known technology may unnecessarily obscure the subject matter of the present disclosure, the detailed description thereof will be omitted. In addition, numbers (eg, first, second, etc.) used in the description process of the specification are only identifiers for distinguishing one component from other components.

In addition, in this specification, when a component is referred to as "connected" or "connected" with another component, the component may be directly connected or directly connected to the other component, but in particular It should be understood that, unless there is a description to the contrary, it may be connected or connected through another element in the middle.

In addition, in the present specification, components expressed as '~ part (unit)', 'module', etc. are two or more components combined into one component, or two or more components for each more subdivided function. may be differentiated into In addition, each of the components to be described below may additionally perform some or all of the functions of other components in addition to the main functions that each component is responsible for, and some of the main functions of each of the components may have different functions. It goes without saying that it may be performed exclusively by the component.

Also, in this specification, an 'image' or 'picture' may indicate a still image, a moving picture composed of a plurality of continuous still images (or frames), or a video.

In addition, in the present specification, a 'deep neural network (DNN)' is a representative example of an artificial neural network model simulating a brain nerve, and is not limited to an artificial neural network model using a specific algorithm.

Also, in the present specification, a 'parameter' is a value used in a calculation process of each layer constituting a neural network, and may include, for example, a weight used when an input value is applied to a predetermined calculation expression. The parameter may be expressed in a matrix form. A parameter is a value set as a result of training, and may be updated through separate training data if necessary.

In addition, in this specification, 'first DNN' means a DNN used for AI downscaling of an image, and 'second DNN' means a DNN used for AI upscaling of an image.

In addition, in the present specification, 'DNN configuration information' includes the aforementioned parameters as information related to elements constituting the DNN. The first DNN or the second DNN may be configured using the DNN configuration information.

In addition, in this specification, the 'original image' refers to an image to be subjected to AI encoding, and the 'first image' refers to an image obtained as a result of AI downscaling of the original image in the AI encoding process. In addition, the 'second image' refers to an image obtained through the first decoding in the AI decoding process, and the 'third image' refers to an image obtained by AI upscaling the second image in the AI decoding process.

In addition, in this specification, 'AI downscaling' refers to a process of reducing the resolution of an image based on AI, and 'first encoding' refers to an encoding process by a frequency transform-based image compression method. In addition, 'first decoding' refers to a decoding process by a frequency transformation-based image restoration method, and 'AI upscaling' refers to a process of increasing the resolution of an image based on AI.

1 is a diagram for explaining an artificial intelligence (AI) encoding process and an AI decoding process according to an embodiment.

As described above, as the resolution of an image rapidly increases, the amount of information processing for encoding/decoding increases. Accordingly, a method for improving encoding and decoding efficiency of an image is required.

As shown in FIG. 1 , according to an embodiment of the present disclosure, a first image 115 is acquired by AI downscaling 110 of an original image 105 having a high resolution. In addition, since the first encoding 120 and the first decoding 130 are performed on the first image 115 of a relatively small resolution, the first encoding 120 and Compared to the case of performing the first decoding 130 , the bit rate may be greatly reduced.

1 , in an embodiment, in the AI encoding process, the original image 105 is AI downscaled 110 to obtain a first image 115 , and the first image 115 is first Encoding (120). In the AI decoding process, AI encoding data including AI data and image data obtained as a result of AI encoding is received, a second image 135 is obtained through the first decoding 130, and the second image 135 is A third image 145 is obtained by AI upscaling 140 .

Looking at the AI encoding process in more detail, when the original image 105 is input, the original image 105 is AI downscaled 110 to obtain the first image 115 of a predetermined resolution and/or a predetermined image quality. At this time, the AI downscaling 110 is performed based on AI, the AI for the AI downscaling 110 must be trained in connection with the AI for the AI upscaling 140 of the second image 135 (joint trained) do. Because, when the AI for the AI downscaling 110 and the AI for the AI upscaling 140 are separately trained, between the original image 105 that is the AI encoding target and the third image 145 restored through AI decoding because the difference between

In an embodiment of the present disclosure, AI data may be used in order to maintain this linkage in the AI encoding process and the AI decoding process. Therefore, the AI data obtained through the AI encoding process should include information indicating the upscale target, and in the AI decoding process, the second image 135 is AI upscaled ( 140) should be done.

AI for AI downscaling 110 and AI for AI upscaling 140 may be implemented as a deep neural network (DNN). As will be described later with reference to FIG. 11 , since the first DNN and the second DNN are jointly trained through sharing of loss information under a predetermined target, the AI encoding apparatus uses target information used when the first DNN and the second DNN are jointly trained. is provided to the AI decoding apparatus, and the AI decoding apparatus may upscale the AI to the image quality and/or resolution targeting the second image 135 based on the received target information.

When the first encoding 120 and the first decoding 130 shown in FIG. 1 are described in detail, the first image 115 AI downscaled 110 from the original image 105 is the first encoding 120 . can reduce the amount of information. The first encoding 120 includes a process of predicting the first image 115 to generate prediction data, a process of generating residual data corresponding to a difference between the first image 115 and the prediction data, and a spatial domain component. It may include a process of transforming the residual data into a frequency domain component, a process of quantizing the residual data transformed into a frequency domain component, and a process of entropy encoding the quantized residual data. Such a first encoding process 120 is MPEG-2, H.264 Advanced Video Coding (AVC), MPEG-4, High Efficiency Video Coding (HEVC), VC-1, VP8, VP9 and AV1 (AOMedia Video 1). It can be implemented through one of the image compression methods using frequency transformation, etc.

The second image 135 corresponding to the first image 115 may be restored through the first decoding 130 of image data. The first decoding 130 includes a process of generating quantized residual data by entropy-decoding image data, a process of inverse quantizing the quantized residual data, a process of transforming the residual data of a frequency domain component into a spatial domain component, a process of predicting data It may include a process of generating , and a process of reconstructing the second image 135 using prediction data and residual data. The first decoding 130 process as described above is an image compression using frequency conversion such as MPEG-2, H.264, MPEG-4, HEVC, VC-1, VP8, VP9 and AV1 used in the first encoding 120 process. It may be implemented through an image restoration method corresponding to one of the methods.

The AI encoded data obtained through the AI encoding process may include image data obtained as a result of the first encoding 120 of the first image 115 and AI data related to the AI downscale 110 of the original image 105. can The image data may be used in the first decoding 130 process, and the AI data may be used in the AI upscaling 140 process.

The image data may be transmitted in the form of a bitstream. The image data may include data obtained based on pixel values in the first image 115 , for example, residual data that is a difference between the first image 115 and prediction data of the first image 115 . . Also, the image data includes information used in the process of the first encoding 120 of the first image 115 . For example, the image data includes prediction mode information used for the first encoding 120 of the first image 115 , motion information, and quantization parameter related information used in the first encoding 120 , etc. can do. The image data is an image compression method used in the first encoding 120 process among image compression methods using frequency conversion, such as MPEG-2, H.264 AVC, MPEG-4, HEVC, VC-1, VP8, VP9, and AV1. It may be generated according to a rule of, for example, a syntax.

AI data is used for AI upscaling 140 based on the second DNN. As described above, since the first DNN and the second DNN are jointly trained, the AI data includes information enabling the accurate AI upscaling 140 of the second image 135 through the second DNN to be performed. . In the AI decoding process, the AI upscaling 140 may be performed to a resolution and/or image quality targeting the second image 135 based on the AI data.

AI data may be transmitted together with image data in the form of a bitstream. According to an embodiment, AI data may be transmitted separately from image data in the form of frames or packets. Alternatively, according to an embodiment, AI data may be transmitted while being included in image data. The image data and AI data may be transmitted through the same network or different networks.

2 is a block diagram illustrating a configuration of an AI decoding apparatus 200 according to an embodiment.

Referring to FIG. 2 , the AI decoding apparatus 200 according to an embodiment includes a receiving unit 210 and an AI decoding unit 230 . The AI decoding unit 230 may include a parsing unit 232 , a first decoding unit 234 , an AI upscaling unit 236 , and an AI setting unit 238 .

Although the receiver 210 and the AI decoder 230 are illustrated as separate devices in FIG. 2 , the receiver 210 and the AI decoder 230 may be implemented through one processor. In this case, the receiving unit 210 and the AI decoding unit 230 may be implemented as a dedicated processor, and a general-purpose processor such as an application processor (AP), a central processing unit (CPU), or a graphic processing unit (GPU) and S/W It may be implemented through a combination of In addition, the dedicated processor may include a memory for implementing an embodiment of the present disclosure or a memory processing unit for using an external memory.

The receiver 210 and the AI decoder 230 may include a plurality of processors. In this case, it may be implemented as a combination of dedicated processors, or may be implemented through a combination of S/W with a plurality of general-purpose processors such as an AP, CPU, or GPU. In one embodiment, the receiving unit 210 is implemented as a first processor, the first decoding unit 234 is implemented as a second processor different from the first processor, the parsing unit 232, the AI upscaling unit 236 and the AI setting unit 238 may be implemented as a third processor different from the first processor and the second processor.

The receiving unit 210 receives AI encoded data obtained as a result of AI encoding. As an example, the AI-encoded data may be a video file having a file format such as mp4 or mov.

The receiver 210 may receive AI-encoded data transmitted through a network. The receiving unit 210 outputs the AI encoded data to the AI decoding unit 230 .

In one embodiment, the AI-encoded data is a hard disk, magnetic media such as floppy disks and magnetic tapes, optical recording media such as CD-ROMs and DVDs, magneto-optical media such as floppy disks. It may be obtained from a data storage medium including an optical medium).

The parsing unit 232 parses the AI encoded data and transmits image data generated as a result of the first encoding of the first image 115 to the first decoder 234 , and transmits the AI data to the AI setting unit 238 . transmit

In an embodiment, the parsing unit 232 may parse the image data and the AI data separately included in the AI encoded data. The parsing unit 232 may read the header in the AI-encoded data to distinguish the AI data and the image data included in the AI-encoded data. In one example, AI data may be included in a Vendor Specific InfoFrame (VSIF) in the HDMI stream. The structure of AI encoded data including AI data and image data separated from each other will be described later with reference to FIG. 9 .

In another embodiment, the parsing unit 232 parses the image data from the AI encoded data, extracts the AI data from the image data, transmits the AI data to the AI setting unit 238, and first decodes the remaining image data. may be transmitted to the unit 234 . That is, AI data may be included in image data. For example, AI data may be included in supplemental enhancement information (SEI), which is an additional information area of a bitstream corresponding to image data. The structure of AI-encoded data including image data including AI data will be described later with reference to FIG. 10 .

In another embodiment, the parsing unit 232 divides the bitstream corresponding to the image data into a bitstream to be processed by the first decoding unit 234 and a bitstream corresponding to AI data, and divides each divided bitstream It may output to the first decoding unit 234 and the AI setting unit 238 .

The parsing unit 232 obtains image data included in the AI encoded data through a predetermined codec (eg, MPEG-2, H.264, MPEG-4, HEVC, VC-1, VP8, VP9, or AV1). It can also be confirmed that it is the image data that has been processed. In this case, the corresponding information may be transmitted to the first decoder 234 so that the image data can be processed by the identified codec.

The first decoder 234 reconstructs the second image 135 corresponding to the first image 115 based on the image data received from the parser 232 . The second image 135 obtained by the first decoder 234 is provided to the AI upscaler 236 .

According to an embodiment, first decoding-related information such as prediction mode information, motion information, and quantization parameter information may be provided from the first decoding unit 234 to the AI setting unit 238 . The first decoding related information may be used to obtain DNN configuration information.

The AI data provided to the AI setting unit 238 includes information enabling AI upscaling of the second image 135 . In this case, the upscale target of the second image 135 should correspond to the downscale target of the first DNN. Therefore, the AI data should include information that can confirm the downscale target of the first DNN.

Specifically exemplifying the information included in the AI data, there are information about a difference between the resolution of the original image 105 and the resolution of the first image 115 , and information related to the first image 115 .

The difference information may be expressed as information about a resolution conversion degree of the first image 115 compared to the original image 105 (eg, resolution conversion rate information). And, since the resolution of the first image 115 is known through the resolution of the restored second image 135 and the degree of resolution conversion can be confirmed through this, the difference information can be expressed only with the resolution information of the original image 105 . may be Here, the resolution information may be expressed as a horizontal/vertical screen size, or it may be expressed as a ratio (16:9, 4:3, etc.) and a size of one axis. In addition, when there is preset resolution information, it may be expressed in the form of an index or a flag.

The information related to the first image 115 is used for the resolution of the first image 115 , the bit rate of image data obtained as a result of the first encoding of the first image 115 , and the first encoding of the first image 115 . information on at least one of the specified codec types may be included.

The AI setting unit 238 may determine an upscale target of the second image 135 based on at least one of difference information included in the AI data and information related to the first image 115 . The upscaling target may indicate, for example, to which resolution the second image 135 should be upscaled.

When the upscaling target is determined, the AI upscaling unit 236 AI upscales the second image 135 through the second DNN to obtain a third image 145 corresponding to the upscaling target.

Prior to explaining a method for the AI setting unit 238 to determine an upscaling target based on AI data, an AI upscaling process through the second DNN will be described with reference to FIGS. 3 and 4 .

3 is an exemplary diagram illustrating a second DNN 300 for AI upscaling of a second image 135, and FIG. 4 is a convolution operation in the first convolution layer 310 shown in FIG. is showing

3 , the second image 135 is input to the first convolutional layer 310 . 3X3X4 displayed in the first convolution layer 310 shown in FIG. 3 exemplifies convolution processing on one input image using four filter kernels having a size of 3×3. As a result of the convolution process, four feature maps are generated by four filter kernels. Each feature map represents unique characteristics of the second image 135 . For example, each feature map may represent a vertical direction characteristic, a horizontal direction characteristic, or an edge characteristic of the second image 135 .

A convolution operation in the first convolution layer 310 will be described in detail with reference to FIG. 4 .

Through the multiplication operation and addition operation between the parameters of the filter kernel 430 having a size of 3 X 3 used in the first convolution layer 310 and the pixel values in the second image 135 corresponding thereto, one A feature map 450 may be generated. Since four filter kernels are used in the first convolution layer 310 , four feature maps may be generated through a convolution operation process using the four filter kernels.

In FIG. 4 , I1 to I49 displayed on the second image 135 indicate pixels of the second image 135 , and F1 to F9 displayed on the filter kernel 430 indicate parameters of the filter kernel 430 . Also, M1 to M9 displayed in the feature map 450 indicate samples of the feature map 450 .

4 illustrates that the second image 135 includes 49 pixels, but this is only an example, and when the second image 135 has a resolution of 4K, for example, 3840 X 2160 pixels It may contain pixels.

In the convolution operation process, pixel values of I1, I2, I3, I8, I9, I10, I15, I16, and I17 of the second image 135 and F1, F2, F3, F4, and F5 of the filter kernel 430 are respectively , F6, F7, F8, and F9 may each be multiplied, and a value obtained by combining (eg, addition operation) result values of the multiplication operation may be assigned as the value of M1 of the feature map 450 . If the stride of the convolution operation is 2, pixel values of I3, I4, I5, I10, I11, I12, I17, I18, and I19 of the second image 135 and F1 and F2 of the filter kernel 430, respectively , F3, F4, F5, F6, F7, F8, and F9 may each be multiplied, and a value obtained by combining result values of the multiplication operation may be assigned as the value of M2 of the feature map 450 .

A convolution operation between the pixel values in the second image 135 and the parameters of the filter kernel 430 is performed while the filter kernel 430 moves along the stride until the last pixel of the second image 135 is reached. By being performed, a feature map 450 having a predetermined size may be obtained.

According to the present disclosure, parameters of the second DNN through joint training of the first DNN and the second DNN, for example, parameters of the filter kernel used in convolutional layers of the second DNN (eg, filter The values of F1, F2, F3, F4, F5, F6, F7, F8, and F9 of the kernel 430 may be optimized. The AI setting unit 238 determines an upscale target corresponding to the downscale target of the first DNN based on the AI data, and uses parameters corresponding to the determined upscale target in the convolution layers of the second DNN. It can be determined by parameters of the filter kernel.

The convolution layers included in the first DNN and the second DNN may be processed according to the convolution operation process described with reference to FIG. 4 , but the convolution operation process described in FIG. 4 is only an example, and the it is not

Referring again to FIG. 3 , the feature maps output from the first convolutional layer 310 are input to the first activation layer 320 .

The first activation layer 320 may provide a non-linear characteristic to each feature map. The first activation layer 320 may include a sigmoid function, a Tanh function, a Rectified Linear Unit (ReLU) function, and the like, but is not limited thereto.

Giving the nonlinear characteristic in the first activation layer 320 means changing and outputting some sample values of the feature map, which is an output of the first convolution layer 310 . At this time, the change is performed by applying a non-linear characteristic.

The first activation layer 320 determines whether to transfer sample values of the feature maps output from the first convolution layer 310 to the second convolution layer 330 . For example, some sample values among the sample values of the feature maps are activated by the first activation layer 320 and transmitted to the second convolution layer 330 , and some sample values are activated by the first activation layer 320 . It is deactivated and is not transmitted to the second convolutional layer 330 . The intrinsic characteristic of the second image 135 indicated by the feature maps is emphasized by the first activation layer 320 .

The feature maps 325 output from the first activation layer 320 are input to the second convolution layer 330 . Any one of the feature maps 325 shown in FIG. 3 is a result of processing the feature map 450 described with reference to FIG. 4 in the first activation layer 320 .

3X3X4 displayed in the second convolution layer 330 exemplifies convolution processing on the input feature maps 325 using four filter kernels having a size of 3×3. The output of the second convolutional layer 330 is input to the second activation layer 340 . The second activation layer 340 may provide a non-linear characteristic to the input data.

The feature maps 345 output from the second activation layer 340 are input to the third convolution layer 350 . 3X3X1 displayed in the third convolution layer 350 shown in FIG. 3 exemplifies convolution processing to generate one output image using one filter kernel having a size of 3×3. The third convolution layer 350 is a layer for outputting a final image and generates one output using one filter kernel. According to the example of the present disclosure, the third convolution layer 350 may output the third image 145 through a convolution operation.

DNN setting information indicating the number of filter kernels of the first convolutional layer 310, the second convolutional layer 330, and the third convolutional layer 350 of the second DNN 300, parameters of the filter kernel, etc. As will be described later, there may be a plurality of pieces, and a plurality of pieces of DNN configuration information should be associated with a plurality of pieces of DNN configuration information of the first DNN. The association between the plurality of DNN configuration information of the second DNN and the plurality of DNN configuration information of the first DNN may be implemented through association learning of the first DNN and the second DNN.

3 shows that the second DNN 300 includes three convolutional layers 310 , 330 , 350 and two activation layers 320 , 340 , but this is only an example, and the implementation Accordingly, the number of convolutional layers and activation layers may be variously changed. Also, according to an embodiment, the second DNN 300 may be implemented through a recurrent neural network (RNN). In this case, it means changing the CNN structure of the second DNN 300 according to the example of the present disclosure to the RNN structure.

In an embodiment, the AI upscaling unit 236 may include at least one arithmetic logic unit (ALU) for the above-described convolution operation and operation of the activation layer. The ALU may be implemented as a processor. For the convolution operation, the ALU includes a multiplier that performs a multiplication operation between the sample values of the feature map output from the second image 135 or the previous layer and the sample values of the filter kernel, and an adder that adds the result values of the multiplication. can In addition, for the calculation of the activation layer, the ALU is a multiplier that multiplies an input sample value by a weight used in a predetermined sigmoid function, a Tanh function, or a ReLU function, and compares the multiplication result with a predetermined value to calculate the input sample value It may include a comparator that determines whether to transfer to the next layer.

Hereinafter, a method in which the AI setting unit 238 determines the upscaling target and the AI upscaling unit 236 AI upscales the second image 135 according to the upscaling target will be described.

In an embodiment, the AI setting unit 238 may store a plurality of DNN setting information settable in the second DNN.

Here, the DNN configuration information may include information on at least one of the number of convolutional layers included in the second DNN, the number of filter kernels for each convolutional layer, and parameters of each filter kernel.

The plurality of DNN configuration information may each correspond to various upscale targets, and the second DNN may operate based on the DNN configuration information corresponding to a specific upscale target. The second DNN may have a different structure according to the DNN configuration information. For example, the second DNN may include three convolutional layers according to certain DNN configuration information, and the second DNN may include four convolutional layers according to other DNN configuration information.

In an embodiment, the DNN configuration information may include only parameters of a filter kernel used in the second DNN. In this case, the structure of the second DNN is not changed, but only the parameters of the internal filter kernel may be changed according to the DNN configuration information.

The AI setting unit 238 may acquire DNN setting information for AI upscaling of the second image 135 among a plurality of DNN setting information. Each of the plurality of DNN configuration information used herein is information for acquiring the third image 145 of a predetermined resolution and/or predetermined image quality, and is trained in association with the first DNN.

For example, the DNN setting information of any one of the plurality of DNN setting information is a third image 145 having a resolution twice that of the second image 135, for example, a second image of 2K (2048 * 1080). It may include information for acquiring the third image 145 of 4K (4096*2160) that is twice as large as the 2nd image 135 , and the other DNN setting information is 4 more than the resolution of the second image 135 . Information for obtaining a third image 145 of twice the resolution, for example, a third image 145 of 8K (8192 * 4320) that is 4 times larger than the second image 135 of 2K (2048 * 1080) may include

Each of the plurality of DNN setting information is created in association with the DNN setting information of the first DNN of the AI encoding device 700, and the AI setting unit 238 is set at an enlargement ratio corresponding to the reduction rate of the DNN setting information of the first DNN. Accordingly, one piece of DNN configuration information among a plurality of DNN configuration information is acquired. To this end, the AI setting unit 238 should check the information of the first DNN. In order for the AI setting unit 238 to check the information of the first DNN, the AI decoding apparatus 200 according to an embodiment receives AI data including the information of the first DNN from the AI encoding apparatus 700 .

In other words, the AI setting unit 238 uses the information received from the AI encoding device 700 to identify information targeted by the DNN setting information of the first DNN used to obtain the first image 115 , and , it is possible to obtain the DNN configuration information of the second DNN trained in association with it.

When DNN setting information for AI upscaling of the second image 135 is obtained among a plurality of DNN setting information, the DNN setting information is transmitted to the AI upscaling unit 236, and a second DNN operating according to the DNN setting information. Based on the input data may be processed.

For example, when any one DNN configuration information is obtained, the AI upscaling unit 236 includes the first convolutional layer 310 and the second convolutional layer 330 of the second DNN 300 shown in FIG. 3 . ) and the third convolutional layer 350, the number of filter kernels included in each layer and parameters of the filter kernels are set to values included in the obtained DNN configuration information.

Specifically, the AI upscaling unit 236 determines that the parameters of the filter kernel of 3 X 3 used in any one convolution layer of the second DNN shown in FIG. 4 are {1, 1, 1, 1, 1, 1 , 1, 1, 1}, and when there is a change in DNN configuration information, the parameters of the filter kernel are changed to {2, 2, 2, 2, 2, 2, 2, 2, 2, parameters included in the changed DNN configuration information. } can be replaced.

The AI setting unit 238 may acquire DNN setting information for upscaling the second image 135 among a plurality of DNN setting information based on the information included in the AI data, which is used to obtain the DNN setting information. AI data will be explained in detail.

In an embodiment, the AI setting unit 238 may obtain DNN setting information for upscaling the second image 135 among a plurality of DNN setting information, based on the difference information included in the AI data. For example, based on the difference information, the resolution of the original image 105 (eg, 4K (4096 * 2160)) is higher than the resolution of the first image 115 (eg, 2K (2048 * 1080)). When it is confirmed that the size is 2 times larger, the AI setting unit 238 may obtain DNN setting information capable of increasing the resolution of the second image 135 by 2 times.

In another embodiment, the AI setting unit 238 is based on the information related to the first image 115 included in the AI data, DNN setting information for AI upscaling the second image 135 among a plurality of DNN setting information. can be obtained. The AI setting unit 238 may determine a mapping relationship between the image related information and the DNN setting information in advance, and obtain DNN setting information mapped to the first image 115 related information.

5 is an exemplary diagram illustrating a mapping relationship between various pieces of image-related information and various pieces of DNN configuration information.

5, it can be seen that the AI encoding/AI decoding process according to an embodiment of the present disclosure does not only consider a change in resolution. As shown in FIG. 5, DNN setting information considering resolutions such as SD, HD, Full HD, bitrates such as 10Mbps, 15Mbps, and 20Mbps, and codec information such as AV1, H.264, and HEVC individually or all selection can be made. For this consideration, training in consideration of each element in the AI training process should be performed in connection with the encoding and decoding processes (see FIG. 11 ).

Accordingly, as shown in FIG. 5 according to the training content, when a plurality of DNN setting information is provided based on image-related information including a codec type and image resolution, the first image ( 115) DNN setting information for AI upscaling of the second image 135 may be acquired based on the related information.

That is, the AI setting unit 238 matches the image-related information shown on the left side of the table shown in FIG. 5 with the DNN setting information on the right side of the table, so that the DNN setting information according to the image-related information can be used.

5 , from the information related to the first image 115 , the resolution of the first image 115 is SD, and the bit rate of image data obtained as a result of the first encoding of the first image 115 is 10Mbps, , when it is confirmed that the first image 115 is first coded with the AV1 codec, the AI setting unit 238 may acquire A DNN setting information among a plurality of DNN setting information.

In addition, from the information related to the first image 115 , the resolution of the first image 115 is HD, the bit rate of image data obtained as a result of the first encoding is 15 Mbps, and the first image 115 is converted to the H.264 codec. When it is confirmed that the first encoding is performed, the AI setting unit 238 may acquire B DNN setting information among a plurality of DNN setting information.

In addition, from the information related to the first image 115 , the resolution of the first image 115 is Full HD, the bit rate of image data obtained as a result of the first encoding of the first image 115 is 20 Mbps, and the first image ( 115) is confirmed to be first encoded by the HEVC codec, the AI setting unit 238 obtains C DNN setting information among a plurality of DNN setting information, the resolution of the first image 115 is Full HD, and the first When the bit rate of the image data obtained as a result of the first encoding of the image 115 is 15 Mbps and it is confirmed that the first image 115 is first encoded with the HEVC codec, the AI setting unit 238 sets a plurality of DNN setting information Among them, DNN configuration information can be obtained. Either one of the C DNN setting information and the D DNN setting information is selected according to whether the bit rate of the image data obtained as a result of the first encoding of the first image 115 is 20 Mbps or 15 Mbps. When the first image 115 of the same resolution is first encoded with the same codec, the fact that the bit rates of the image data are different from each other means that the image quality of the reconstructed images is different from each other. Therefore, the first DNN and the second DNN may be jointly trained based on a predetermined image quality, and accordingly, the AI setting unit 238 sets the DNN according to the bit rate of the image data indicating the image quality of the second image 135 . information can be obtained.

In another embodiment, the AI setting unit 238 relates to the information (prediction mode information, motion information, quantization parameter information, etc.) provided from the first decoder 234 and the first image 115 included in the AI data. DNN configuration information for AI upscaling of the second image 135 among a plurality of pieces of DNN configuration information may be acquired in consideration of all information. For example, the AI setting unit 238 receives quantization parameter information used in the first encoding process of the first image 115 from the first decoder 234 , and receives the quantization parameter information from the AI data of the first image 115 . A bit rate of image data obtained as a result of encoding may be checked, and DNN setting information corresponding to a quantization parameter and a bit rate may be obtained. Even at the same bit rate, there may be a difference in the quality of the reconstructed image depending on the complexity of the image, and the bit rate is a value representative of the entire first image 115 to be first encoded, and the bit rate of each frame in the first image 115 is The picture quality may be different. Accordingly, when the prediction mode information, motion information, and/or quantization parameters obtainable from the first decoder 234 for each frame are considered together, the DNN setting more suitable for the second image 135 compared to using only AI data information can be obtained.

Also, depending on the implementation, the AI data may include an identifier of mutually agreed DNN configuration information. The identifier of the DNN setting information is an upscale target corresponding to the downscale target of the first DNN, and the DNN setting information trained in association between the first DNN and the second DNN so that the second image 135 can be AI upscaled. It is information for distinguishing pairs of . The AI setting unit 238 obtains the identifier of the DNN setting information included in the AI data, and then obtains the DNN setting information corresponding to the identifier of the DNN setting information, and the AI upscaling unit 236 receives the corresponding DNN setting information. The second image 135 may be upscaled using AI. For example, an identifier indicating each of a plurality of pieces of DNN configuration information configurable in the first DNN and an identifier indicating each of a plurality of pieces of DNN configuration information configurable in the second DNN may be predefined. In this case, the same identifier may be designated for a pair of DNN configuration information configurable to each of the first DNN and the second DNN. The AI data may include an identifier of DNN setting information set in the first DNN for AI downscaling of the original image 105 . The AI setting unit 238 that has received the AI data acquires the DNN setting information indicated by the identifier included in the AI data among a plurality of DNN setting information, and the AI upscaling unit 236 uses the corresponding DNN setting information to obtain the second The image 135 may be upscaled by AI.

Also, according to implementation, AI data may include DNN configuration information. The AI setting unit 238 may obtain DNN setting information included in AI data, and the AI upscaling unit 236 may AI upscale the second image 135 using the corresponding DNN setting information.

According to the embodiment, when information constituting the DNN setting information (eg, the number of convolution layers, the number of filter kernels for each convolutional layer, parameters of each filter kernel, etc.) is stored in the form of a lookup table, The AI setting unit 238 obtains DNN setting information by combining some selected among the lookup table values based on information included in the AI data, and the AI upscaling unit 236 uses the corresponding DNN setting information to obtain the second image (135) may be AI upscaled.

According to an embodiment, when the structure of the DNN corresponding to the upscale target is determined, the AI setting unit 238 may obtain DNN setting information corresponding to the determined structure of the DNN, for example, parameters of the filter kernel.

As described above, the AI setting unit 238 obtains DNN setting information of the second DNN through AI data including information related to the first DNN, and the AI upscaling unit 236 sets the DNN setting information to the corresponding DNN setting information. AI upscales the second image 135 through the second DNN, which may reduce memory usage and computational amount compared to upscaling by directly analyzing the characteristics of the second image 135 .

In an embodiment, when the second image 135 consists of a plurality of frames, the AI setting unit 238 may independently obtain DNN setting information for each predetermined number of frames, or a common DNN for all frames. It is also possible to obtain setting information.

6 is a diagram illustrating a second image 135 composed of a plurality of frames.

As shown in FIG. 6 , the second image 135 may include frames corresponding to t0 to tn.

In one example, the AI setting unit 238 obtains DNN setting information of the second DNN through AI data, and the AI upscaling unit 236 AI frames corresponding to t0 to tn based on the corresponding DNN setting information. can be upscaled. That is, frames corresponding to t0 to tn may be AI upscaled based on common DNN configuration information.

In another example, the AI setting unit 238 obtains 'A' DNN setting information from AI data for some of the frames corresponding to t0 to tn, for example, frames corresponding to t0 to ta. and 'B' DNN configuration information may be obtained from AI data for frames corresponding to ta+1 to tb. Also, the AI setting unit 238 may obtain 'C' DNN setting information from AI data for frames corresponding to tb+1 to tn. In other words, the AI setting unit 238 independently obtains DNN setting information for each group including a predetermined number of frames among the plurality of frames, and the AI upscaling unit 236 independently selects the frames included in each group. AI upscaling is possible with the acquired DNN configuration information.

In another example, the AI setting unit 238 may independently obtain DNN setting information for each frame constituting the second image 135 . For example, when the second image 135 consists of three frames, the AI setting unit 238 obtains DNN setting information in relation to the first frame, and DNN setting information in relation to the second frame. and may acquire DNN configuration information in relation to the third frame. That is, DNN configuration information may be independently obtained for each of the first frame, the second frame, and the third frame. DNN setting information is obtained based on the information (prediction mode information, motion information, quantization parameter information, etc.) provided from the above-described first decoder 234 and information related to the first image 115 included in the AI data According to the method, DNN setting information may be independently obtained for each frame constituting the second image 135 . This is because mode information, quantization parameter information, and the like can be independently determined for each frame constituting the second image 135 .

In another example, the AI data may include information indicating to which frame the DNN configuration information obtained based on the AI data is valid. For example, if the AI data includes information that the DNN setting information is valid until the ta frame, the AI setting unit 238 acquires the DNN setting information based on the AI data, and the AI upscaling unit 236 is AI upscales t0 to ta frames with corresponding DNN configuration information. And, when the information that the DNN setting information is valid until the tn frame is included in other AI data, the AI setting unit 238 acquires the DNN setting information based on the other AI data, and the AI upscaling unit 236 may AI upscale ta+1 to tn frames with the obtained DNN configuration information.

Hereinafter, an AI encoding apparatus 700 for AI encoding of the original image 105 will be described with reference to FIG. 7 .

7 is a block diagram illustrating a configuration of an AI encoding apparatus 700 according to an embodiment.

Referring to FIG. 7 , the AI encoding apparatus 700 may include an AI encoding unit 710 and a transmission unit 730 . The AI encoder 710 may include an AI downscaler 712 , a first encoder 714 , a data processor 716 , and an AI setting unit 718 .

7 illustrates the AI encoder 710 and the transmitter 730 as separate devices, the AI encoder 710 and the transmitter 730 may be implemented through one processor. In this case, it may be implemented as a dedicated processor, or it may be implemented through a combination of an AP or a general-purpose processor such as a CPU or GPU and S/W. In addition, the dedicated processor may include a memory for implementing an embodiment of the present disclosure or a memory processing unit for using an external memory.

Also, the AI encoder 710 and the transmitter 730 may include a plurality of processors. In this case, it may be implemented as a combination of dedicated processors, or may be implemented through a combination of S/W with a plurality of general-purpose processors such as an AP, CPU, or GPU. In an embodiment, the first encoding unit 714 is configured as a first processor, and the AI downscale unit 712 , the data processing unit 716 , and the AI setting unit 718 are configured as a second processor different from the first processor. In the implementation, the transmitter 730 may be implemented as a third processor different from the first processor and the second processor.

The AI encoder 710 performs AI downscaling of the original image 105 and the first encoding of the first image 115 , and transmits the AI encoded data to the transmitter 730 . The transmitter 730 transmits the AI encoded data to the AI decoding apparatus 200 .

The image data includes data obtained as a result of the first encoding of the first image 115 . The image data may include data obtained based on pixel values in the first image 115 , for example, residual data that is a difference between the first image 115 and prediction data of the first image 115 . . Also, the image data includes information used in the first encoding process of the first image 115 . For example, the image data may include prediction mode information used for first encoding the first image 115 , motion information, and quantization parameter related information used for first encoding the first image 115 , etc. .

The AI data includes information enabling the AI upscaling unit 236 to AI upscale the second image 135 to an upscale target corresponding to the downscale target of the first DNN.

In one example, the AI data may include difference information between the original image 105 and the first image 115 .

In an example, the AI data may include information related to the first image 115 . The information related to the first image 115 is used for the resolution of the first image 115 , the bit rate of image data obtained as a result of the first encoding of the first image 115 , and the first encoding of the first image 115 . information on at least one of the specified codec types may be included.

In an embodiment, the AI data may include an identifier of mutually agreed DNN configuration information so that the second image 135 can be AI upscaled to an upscale target corresponding to the downscale target of the first DNN. .

Also, in an embodiment, the AI data may include DNN configuration information settable in the second DNN.

The AI downscaler 712 may acquire the AI downscaled first image 115 from the original image 105 through the first DNN. The AI downscaling unit 712 may AI downscale the original image 105 using the DNN setting information provided from the AI setting unit 718 .

The AI setting unit 718 may determine a downscale target of the original image 105 based on a predetermined criterion.

In order to acquire the first image 115 matching the downscale target, the AI setting unit 718 may store a plurality of settable DNN setting information in the first DNN. The AI setting unit 718 obtains DNN setting information corresponding to the downscale target among a plurality of DNN setting information, and provides the obtained DNN setting information to the AI downscale unit 712 .

Each of the plurality of pieces of DNN configuration information may be trained to obtain the first image 115 having a predetermined resolution and/or a predetermined image quality. For example, the DNN setting information of any one of the plurality of DNN setting information is the first image 115 having a resolution that is 1/2 times smaller than the resolution of the original image 105, for example, 4K (4096*2160) may include information for obtaining the first image 115 of 2K (2048*1080) 1/2 times smaller than the original image 105 of A first image 115 of a resolution of 1/4 times smaller than that of a first image 115, for example, a first image 115 of 2K (2048*1080) which is 1/4 times smaller than an original image 105 of 8K (8192*4320). ) may include information for obtaining

According to the embodiment, when information constituting the DNN setting information (eg, the number of convolution layers, the number of filter kernels for each convolutional layer, parameters of each filter kernel, etc.) is stored in the form of a lookup table, The AI setting unit 718 may obtain DNN setting information by combining some selected among the lookup table values according to the downscale target, and may provide the obtained DNN setting information to the AI downscaling unit 712 .

According to an embodiment, the AI setting unit 718 may determine the structure of the DNN corresponding to the downscale target, and obtain DNN setting information corresponding to the determined structure of the DNN, for example, parameters of the filter kernel.

The plurality of DNN configuration information for AI downscaling of the original image 105 may have an optimized value by performing joint training of the first DNN and the second DNN. Here, each DNN configuration information includes at least one of the number of convolutional layers included in the first DNN, the number of filter kernels for each convolutional layer, and parameters of each filter kernel.

The AI downscaling unit 712 sets the first DNN as the DNN setting information determined for the AI downscaling of the original image 105, and converts the first image 115 of a predetermined resolution and/or a predetermined quality to the first DNN. can be obtained through When DNN setting information for AI downscaling of the original image 105 is obtained among a plurality of DNN setting information, each layer in the first DNN may process input data based on information included in the DNN setting information. .

Hereinafter, a method for the AI setting unit 718 to determine the downscale target will be described. The downscale target may indicate, for example, how much the first image 115 with reduced resolution should be obtained from the original image 105 .

The AI setting unit 718 obtains one or more pieces of input information. In an embodiment, the input information includes a target resolution of the first image 115, a target bitrate of the image data, a bitrate type of the image data (eg, a variable bitrate type, a constant bitrate type, or an average bitrate type); Color format to which AI downscale is applied (luminance component, chrominance component, red component, green component, or blue component, etc.), codec type for first encoding of first image 115, compression history information, original image 105 may include at least one of a resolution of , and a type of the original image 105 .

One or more pieces of input information may include information previously stored in the AI encoding apparatus 700 or received from a user.

The AI setting unit 718 controls the operation of the AI downscale unit 712 based on the input information. In an embodiment, the AI setting unit 718 may determine a downscale target according to input information, and provide DNN setting information corresponding to the determined downscale target to the AI downscale unit 712 .

In an embodiment, the AI setting unit 718 transmits at least a portion of the input information to the first encoding unit 714 so that the first encoding unit 714 provides a bitrate of a specific value, a bitrate of a specific type, and a specific codec. may cause the first image 115 to be first encoded.

In an embodiment, the AI setting unit 718 may include a compression rate (eg, a difference in resolution between the original image 105 and the first image 115 , a target bitrate), and a compression quality (eg, a bitrate type). ), compression history information, and a downscale target may be determined based on at least one of the type of the original image 105 .

In an example, the AI setting unit 718 may determine a downscale target based on a preset or a compression rate or compression quality input from a user.

As another example, the AI setting unit 718 may determine the downscale target by using the compression history information stored in the AI encoding apparatus 700 . For example, according to the compression history information available to the AI encoding apparatus 700, the encoding quality or compression rate preferred by the user may be determined, and the downscale target may be determined according to the encoding quality determined based on the compression history information. can For example, the resolution and quality of the first image 115 may be determined according to the encoding quality that has been most frequently used according to the compression history information.

As another example, the AI setting unit 718 sets the encoding quality that has been used more than a predetermined threshold value (eg, average quality of encoding qualities that have been used more than a predetermined threshold value) according to the compression history information. A downscale target may be determined based on it.

As another example, the AI setting unit 718 may determine the downscale target based on the resolution and type (eg, file format) of the original image 105 .

In one embodiment, when the original image 105 is composed of a plurality of frames, the AI setting unit 718 independently acquires DNN setting information for each predetermined number of frames, and uses the independently acquired DNN setting information to AI downscaling. A portion 712 may be provided.

In one example, the AI setting unit 718 may classify frames constituting the original image 105 into a predetermined number of groups, and independently obtain DNN setting information for each group. The same or different DNN configuration information may be obtained for each group. The number of frames included in the groups may be the same or different for each group.

In another example, the AI setting unit 718 may independently determine DNN setting information for each frame constituting the original image 105 . The same or different DNN configuration information may be obtained for each frame.

Hereinafter, an exemplary structure of the first DNN 800 as a basis for AI downscaling will be described.

8 is an exemplary diagram illustrating the first DNN 800 for AI downscaling of the original image 105 .

As shown in FIG. 8 , the original image 105 is input to the first convolutional layer 810 . The first convolutional layer 810 performs convolution processing on the original image 105 using 32 filter kernels having a size of 5×5. The 32 feature maps generated as a result of the convolution process are input to the first activation layer 820 .

The first activation layer 820 may provide non-linear characteristics to 32 feature maps.

The first activation layer 820 determines whether to transfer sample values of the feature maps output from the first convolution layer 810 to the second convolution layer 830 . For example, some sample values among the sample values of the feature maps are activated by the first activation layer 820 and transmitted to the second convolution layer 830 , and some sample values are activated by the first activation layer 820 . It is deactivated and is not transmitted to the second convolutional layer 830 . Information indicated by the feature maps output from the first convolution layer 810 is emphasized by the first activation layer 820 .

The output 825 of the first activation layer 820 is input to the second convolution layer 830 . The second convolution layer 830 performs convolution processing on input data by using 32 filter kernels having a size of 5×5. The 32 feature maps output as a result of the convolution process are input to the second activation layer 840 , and the second activation layer 840 may provide nonlinear characteristics to the 32 feature maps.

The output 845 of the second activation layer 840 is input to the third convolution layer 850 . The third convolution layer 850 performs convolution processing on input data using one filter kernel having a size of 5×5. As a result of the convolution processing, one image may be output from the third convolutional layer 850 . The third convolutional layer 850 is a layer for outputting a final image and obtains one output by using one filter kernel. According to the example of the present disclosure, the third convolution layer 850 may output the first image 115 through the convolution operation result.

DNN setting information indicating the number of filter kernels of the first convolutional layer 810, the second convolutional layer 830, and the third convolutional layer 850 of the first DNN 800, parameters of the filter kernel, etc. There may be a plurality, and the plurality of DNN configuration information should be associated with the plurality of DNN configuration information of the second DNN. The association between the plurality of DNN configuration information of the first DNN and the plurality of DNN configuration information of the second DNN may be implemented through association learning of the first DNN and the second DNN.

8 shows that the first DNN 800 includes three convolutional layers 810 , 830 , 850 and two activation layers 820 , 840 , but this is only an example, and the implementation Accordingly, the number of convolutional layers and activation layers may be variously changed. Also, according to an embodiment, the first DNN 800 may be implemented through a recurrent neural network (RNN). In this case, it means changing the CNN structure of the first DNN 800 according to the example of the present disclosure to the RNN structure.

In an embodiment, the AI downscaling unit 712 may include at least one ALU for a convolution operation and an activation layer operation. The ALU may be implemented as a processor. For the convolution operation, the ALU may include a multiplier that performs a multiplication operation between the sample values of the feature map output from the original image 105 or the previous layer and the sample values of the filter kernel, and an adder that adds the result values of the multiplication. have. In addition, for the calculation of the activation layer, the ALU is a multiplier that multiplies an input sample value by a weight used in a predetermined sigmoid function, a Tanh function, or a ReLU function, and compares the multiplication result with a predetermined value to calculate the input sample value It may include a comparator that determines whether to transfer to the next layer.

Referring back to FIG. 7 , the AI setting unit 718 transmits AI data to the data processing unit 716 . The AI data includes information enabling the AI upscaling unit 236 to AI upscale the second image 135 to an upscale target corresponding to the downscale target of the first DNN.

The first encoder 714 receiving the first image 115 from the AI downscaler 712 first encodes the first image 115 according to the frequency transformation-based image compression method to obtain the first image 115 . ) can reduce the amount of information it has. As a result of the first encoding through a predetermined codec (eg, MPEG-2, H.264, MPEG-4, HEVC, VC-1, VP8, VP9, or AV1), image data is obtained. The image data is obtained according to a rule of a predetermined codec, that is, a syntax. For example, the image data includes residual data that is a difference between the first image 115 and the prediction data of the first image 115 , prediction mode information used to first encode the first image 115 , motion information, and Information related to a quantization parameter used to first encode the first image 115 may be included.

The image data obtained as a result of the first encoding by the first encoder 714 is provided to the data processor 716 .

The data processing unit 716 generates AI encoded data including the image data received from the first encoding unit 714 and the AI data received from the AI setting unit 718 .

In an embodiment, the data processing unit 716 may generate AI-encoded data including image data and AI data in a separate state. As an example, AI data may be included in a Vendor Specific InfoFrame (VSIF) in the HDMI stream.

In another embodiment, the data processing unit 716 may include AI data in image data obtained as a result of the first encoding by the first encoding unit 714 and generate AI encoded data including the corresponding image data. . For example, the data processing unit 716 may generate image data in the form of one bitstream by combining a bitstream corresponding to image data and a bitstream corresponding to AI data. To this end, the data processing unit 716 may represent the AI data as bits having a value of 0 or 1, that is, a bitstream. In an embodiment, the data processing unit 716 may include a bitstream corresponding to AI data in supplemental enhancement information (SEI) that is an additional information area of a bitstream obtained as a result of the first encoding.

The AI-encoded data is transmitted to the transmitter 730 . The transmitter 730 transmits the AI-encoded data obtained as a result of the AI-encoding through a network.

In one embodiment, the AI encoded data is a hard disk, a magnetic medium such as a floppy disk and magnetic tape, an optical recording medium such as CD-ROM and DVD, a magneto-optical medium such as a floppy disk. medium) may be stored in a data storage medium including

9 is a diagram illustrating a structure of AI-encoded data 900 according to an embodiment.

As described above, the AI data 912 and the image data 932 may be separately included in the AI encoded data 900 . Here, the AI-encoded data 900 may be in a container format such as MP4, AVI, MKV, or FLV. The AI-encoded data 900 may include a metadata box 910 and a media data box 930 .

The metadata box 910 includes information about the image data 932 included in the media data box 930 . For example, the metadata box 910 may include information on the type of the first image 115 , the type of codec used to encode the first image 115 , and the playback time of the first image 115 . have. Also, the metadata box 910 may include AI data 912 . The AI data 912 may be encoded according to an encoding method provided by a predetermined container format and stored in the metadata box 910 .

The media data box 930 may include image data 932 generated according to the syntax of a predetermined image compression method.

10 is a diagram illustrating a structure of AI encoded data 1000 according to another embodiment.

Referring to FIG. 10 , AI data 1034 may be included in image data 1032 . The AI-encoded data 1000 may include a metadata box 1010 and a media data box 1030 . When the AI data 1034 is included in the image data 1032 , the metadata box 1010 contains AI Data 1034 may not be included.

The media data box 1030 includes image data 1032 including AI data 1034 . For example, the AI data 1034 may be included in the additional information area of the image data 1032 .

Hereinafter, a method of jointly training the first DNN 800 and the second DNN 300 will be described with reference to FIG. 11 .

11 is a diagram for explaining a method of training the first DNN 800 and the second DNN 300 .

In an embodiment, the AI-encoded original image 105 through the AI encoding process is restored to the third image 145 through the AI decoding process. The third image 145 and the original image 105 obtained as a result of AI decoding In order to maintain the similarity with the AI encoding process and the AI decoding process, correlation is required. That is, information lost in the AI encoding process should be able to be restored in the AI decoding process, and for this purpose, joint training between the first DNN 800 and the second DNN 300 is required.

For accurate AI decoding, it is ultimately necessary to reduce the quality loss information 1130 corresponding to the comparison result between the third training image 1104 and the original training image 1101 shown in FIG. 11 . Accordingly, the quality loss information 1130 is used for both training of the first DNN 800 and the second DNN 300 .

First, the training process shown in FIG. 11 will be described.

In FIG. 11 , an original training image 1101 is an image subject to AI downscale, and a first training image 1102 is an AI downscaled image from the original training image 1101 . it's a video Also, the third training image 1104 is an AI upscaled image from the first training image 1102 .

The original training image 1101 includes a still image or a moving image composed of a plurality of frames. In an embodiment, the original training image 1101 may include a still image or a luminance image extracted from a moving picture including a plurality of frames. Also, according to an embodiment, the original training image 1101 may include a still image or a patch image extracted from a moving image including a plurality of frames. When the original training image 1101 consists of a plurality of frames, the first training image 1102 , the second training image and the third training image 1104 also include a plurality of frames. When a plurality of frames of the original training image 1101 are sequentially input to the first DNN 800 , the first training image 1102 and the second training image are performed through the first DNN 800 and the second DNN 300 . and a plurality of frames of the third training image 1104 may be sequentially acquired.

For joint training of the first DNN 800 and the second DNN 300 , an original training image 1101 is input to the first DNN 800 . The original training image 1101 input to the first DNN 800 is AI downscaled and output as the first training image 1102 , and the first training image 1102 is input to the second DNN 300 . A third training image 1104 is output as a result of AI upscaling for the first training image 1102 .

Referring to FIG. 11 , a first training image 1102 is being input to the second DNN 300 . According to an embodiment, the first training image 1102 is obtained through a first encoding and a first decoding process. A second training image may be input to the second DNN 300 . In order to input the second training image to the second DNN, any one of MPEG-2, H.264, MPEG-4, HEVC, VC-1, VP8, VP9, and AV1 codecs may be used. Specifically, in the first encoding of the first training image 1102 and the first decoding of image data corresponding to the first training image 1102, MPEG-2, H.264, MPEG-4, HEVC, VC-1 , VP8, VP9, and any one codec of AV1 may be used.

Referring to FIG. 11 , the legacy downscaled reduced training image 1103 is obtained from the original training image 1101 separately from the output of the first training image 1102 through the first DNN 800 . Here, the legacy downscale may include at least one of a bilinear scale, a bicubic scale, a lanczos scale, and a stair step scale.

In order to prevent the structural features of the first image 115 from greatly deviate based on the structural features of the original image 105, a reduced training image 1103 that preserves the structural features of the original training image 1101 is acquired. .

Before the training proceeds, the first DNN 800 and the second DNN 300 may be set with predetermined DNN configuration information. As training progresses, structural loss information 1110 , complexity loss information 1120 , and quality loss information 1130 may be determined.

The structural loss information 1110 may be determined based on a comparison result between the reduced training image 1103 and the first training image 1102 . In one example, the structural loss information 1110 may correspond to a difference between the structural information of the reduced training image 1103 and the structural information of the first training image 1102 . The structural information may include various features that can be extracted from the image, such as luminance, contrast, and histogram of the image. The structural loss information 1110 indicates to what extent the structural information of the original training image 1101 is maintained in the first training image 1102 . As the structural loss information 1110 is smaller, the structural information of the first training image 1102 is similar to the structural information of the original training image 1101 .

The complexity loss information 1120 may be determined based on the spatial complexity of the first training image 1102 . In an example, as the spatial complexity, a total variance value of the first training image 1102 may be used. The complexity loss information 1120 is related to a bit rate of image data obtained by first encoding the first training image 1102 . The smaller the complexity loss information 1120 is, the smaller the bit rate of image data is defined.

The quality loss information 1130 may be determined based on a comparison result between the original training image 1101 and the third training image 1104 . The quality loss information 1130 includes an L1-norm value, an L2-norm value, a Structural Similarity (SSIM) value, and a Peak Signal-To (PSNR-HVS) value for the difference between the original training image 1101 and the third training image 1104 . It may include at least one of a Noise Ratio-Human Vision System) value, a Multiscale SSIM (MS-SSIM) value, a Variance Inflation Factor (VIF) value, and a Video Multimethod Assessment Fusion (VMAF) value. The quality loss information 1130 indicates how similar the third training image 1104 is to the original training image 1101 . As the quality loss information 1130 is smaller, the third training image 1104 is more similar to the original training image 1101 .

11, structural loss information 1110, complexity loss information 1120, and quality loss information 1130 are used for training of the first DNN 800, and the quality loss information 1130 is the second DNN ( 300) is used for training. That is, the quality loss information 1130 is used for both training of the first DNN 800 and the second DNN 300 .

The first DNN 800 may update parameters such that the final loss information determined based on the structural loss information 1110 , the complexity loss information 1120 , and the quality loss information 1130 is reduced or minimized. Also, the second DNN 300 may update the parameter so that the quality loss information 1130 is reduced or minimized.

Final loss information for training of the first DNN 800 and the second DNN 300 may be determined as in Equation 1 below.

[Equation 1]

In Equation 1, LossDS represents final loss information to be reduced or minimized for training of the first DNN 800, and LossUS represents final loss information to be reduced or minimized for training of the second DNN 300. indicates. Also, a, b, c, and d may correspond to predetermined weights.

That is, the first DNN 800 updates parameters in a direction in which LossDS of Equation 1 is decreased, and the second DNN 300 updates parameters in a direction in which LossUS is decreased. When the parameters of the first DNN 800 are updated according to the LossDS derived in the training process, the first training image 1102 obtained based on the updated parameters is different from the first training image 1102 in the previous training process. and, accordingly, the third training image 1104 is also different from the third training image 1104 in the previous training process. When the third training image 1104 is different from the third training image 1104 in the previous training process, the quality loss information 1130 is also newly determined, and accordingly, the second DNN 300 updates the parameters. When the quality loss information 1130 is newly determined, since LossDS is also newly determined, the first DNN 800 updates parameters according to the newly determined LossDS. That is, the parameter update of the first DNN 800 causes the parameter update of the second DNN 300 , and the parameter update of the second DNN 300 causes the parameter update of the first DNN 800 . In other words, since the first DNN 800 and the second DNN 300 are jointly trained through sharing of the quality loss information 1130 , the parameters of the first DNN 800 and the parameters of the second DNN 300 are They can be optimized in relation to each other.

Referring to Equation 1, it can be seen that LossUS is determined according to quality loss information 1130, but this is an example, and LossUS is at least one of structural loss information 1110 and complexity loss information 1120, It may be determined based on the quality loss information 1130 .

Previously, it has been described that the AI setting unit 238 of the AI decoding apparatus 200 and the AI setting unit 718 of the AI encoding apparatus 700 store a plurality of DNN setting information, the AI setting unit 238 and the AI A method of training each of the plurality of DNN configuration information stored in the setting unit 718 will be described.

As described in relation to Equation 1, in the case of the first DNN 800, the degree of similarity between the structural information of the first training image 1102 and the structural information of the original training image 1101 (structural loss information 1110) ), the bit rate (complexity loss information 1120) of the image data obtained as a result of the first encoding of the first training image 1102 and the difference between the third training image 1104 and the original training image 1101 (quality loss) The parameter is updated in consideration of the information 1130).

In detail, the first training image 1102 similar to the structural information of the original training image 1101 and having a small bit rate of image data obtained when the first encoding is performed can be obtained, and at the same time, the first training image The parameters of the first DNN 800 may be updated so that the second DNN 300 that AI upscales 1102 may acquire a third training image 1104 similar to the original training image 1101 .

By adjusting the weights of a, b, and c in Equation 1, the directions in which the parameters of the first DNN 800 are optimized are different. For example, when the weight of b is determined to be high, the parameter of the first DNN 800 may be updated with more importance on lowering the bit rate than the quality of the third training image 1104 . In addition, when the weight of c is determined to be high, it is more important to increase the quality of the third training image 1104 than to increase the bit rate or to maintain the structural information of the original training image 1101. 1 A parameter of the DNN 800 may be updated.

Also, the direction in which parameters of the first DNN 800 are optimized may be different according to the type of a codec used to first encode the first training image 1102 . This is because, depending on the type of codec, the second training image to be input to the second DNN 300 may vary.

That is, the parameters of the first DNN 800 and the parameters of the second DNN 300 are linked based on the weight a, the weight b, the weight c, and the type of the codec for the first encoding of the first training image 1102 . so it can be updated. Therefore, when each of the weight a, the weight b, and the weight c is determined as a predetermined value and the type of the codec is determined as the predetermined type, the first DNN 800 and the second DNN 300 are trained, they are linked to each other. Optimized parameters of the first DNN 800 and parameters of the second DNN 300 may be determined.

Then, when the first DNN 800 and the second DNN 300 are trained after changing the weight a, the weight b, the weight c, and the type of the codec, the parameters of the first DNN 800 are optimized in connection with each other. and parameters of the second DNN 300 may be determined. In other words, when the first DNN 800 and the second DNN 300 are trained while changing the respective values of the weight a, the weight b, the weight c, and the type of the codec, the plurality of DNN setting information trained in connection with each other is transmitted to the first It may be determined from the DNN 800 and the second DNN 300 .

As described above with reference to FIG. 5 , the plurality of DNN configuration information of the first DNN 800 and the second DNN 300 may be mapped to the first image related information. To establish such a mapping relationship, a first training image 1102 output from the first DNN 800 is first encoded with a specific codec according to a specific bit rate, and a bitstream obtained as a result of the first encoding is first decoded. The obtained second training image may be input to the second DNN 300 . That is, after setting the environment so that the first training image 1102 of a specific resolution is first encoded at a specific bit rate by a specific codec, by training the first DNN 800 and the second DNN 300 , the first DNN setting mapped to the resolution of the training image 1102, the type of codec used for the first encoding of the first training image 1102, and the bitrate of the bitstream obtained as a result of the first encoding of the first training image 1102 The information pair can be determined. The resolution of the first training image 1102, the type of codec used for the first encoding of the first training image 1102, and the bitrate of the bitstream obtained according to the first encoding of the first training image 1102 are varied. , a mapping relationship between a plurality of DNN configuration information of the first DNN 800 and the second DNN 300 and the first image related information may be determined.

12 is a diagram for explaining a training process of the first DNN 800 and the second DNN 300 by the training device 1200 .

The training of the first DNN 800 and the second DNN 300 described with reference to FIG. 11 may be performed by the training apparatus 1200 . The training device 1200 includes a first DNN 800 and a second DNN 300 . The training device 1200 may be, for example, the AI encoding device 700 or a separate server. The DNN setting information of the second DNN 300 obtained as a result of training is stored in the AI decoding apparatus 200 .

Referring to FIG. 12 , the training apparatus 1200 initially sets DNN configuration information of the first DNN 800 and the second DNN 300 ( S1240 , S1245 ). Accordingly, the first DNN 800 and the second DNN 300 may operate according to predetermined DNN configuration information. The DNN setting information includes at least the number of convolution layers included in the first DNN 800 and the second DNN 300, the number of filter kernels for each convolutional layer, the size of filter kernels for each convolutional layer, and parameters of each filter kernel. It can contain information about one.

The training apparatus 1200 inputs the original training image 1101 into the first DNN 800 (S1250). The original training image 1101 may include at least one frame constituting a still image or a moving image.

The first DNN 800 processes the original training image 1101 according to the initially set DNN setting information, and outputs the AI downscaled first training image 1102 from the original training image 1101 ( S1255 ). 12 shows that the first training image 1102 output from the first DNN 800 is directly input to the second DNN 300 , but the first training image 1102 output from the first DNN 800 . ) may be input to the second DNN 300 by the training device 1200 . Also, the training apparatus 1200 may first encode and first decode the first training image 1102 using a predetermined codec, and then input the second training image to the second DNN 300 .

The second DNN 300 processes the first training image 1102 or the second training image according to the initially set DNN setting information, and the AI upscaled third from the first training image 1102 or the second training image. A training image 1104 is output (S1260).

The training apparatus 1200 calculates complexity loss information 1120 based on the first training image 1102 (S1265).

The training apparatus 1200 calculates structural loss information 1110 by comparing the reduced training image 1103 with the first training image 1102 ( S1270 ).

The training apparatus 1200 calculates quality loss information 1130 by comparing the original training image 1101 and the third training image 1104 (S1275).

The first DNN 800 updates the initially set DNN configuration information through a back propagation process based on the final loss information (S1280). The training apparatus 1200 may calculate final loss information for training the first DNN 800 based on the complexity loss information 1120 , the structural loss information 1110 , and the quality loss information 1130 .

The second DNN 300 updates the initially set DNN configuration information through a reverse transcription process based on quality loss information or final loss information (S1285). The training apparatus 1200 may calculate final loss information for training of the second DNN 300 based on the quality loss information 1130 .

Thereafter, the training apparatus 1200, the first DNN 800, and the second DNN 300 update the DNN configuration information while repeating processes S1250 to S1285 until the final loss information is minimized. At this time, during each iteration process, the first DNN 800 and the second DNN 300 operate according to the DNN configuration information updated in the previous process.

Table 1 below shows the effects of AI encoding and AI decoding of the original image 105 and encoding and decoding of the original image 105 using HEVC according to an embodiment of the present disclosure.

[Table 1]

As can be seen from Table 1, the subjective picture quality when AI encoding and AI decoding of contents consisting of 300 frames of 8K resolution according to an embodiment of the present disclosure is higher than the subjective picture quality when encoding and decoding using HEVC. , it can be seen that the bit rate is reduced by more than 50%.

Hereinafter, with reference to FIGS. 13 to 27 , an AI encoding apparatus and an AI decoding apparatus according to an embodiment process AI data through a first encoder (and data processor) and a first decoder (and a parser) A method is disclosed.

13A is a block diagram illustrating a configuration of an AI decoding apparatus 1300 according to an embodiment.

Referring to FIG. 13 , the AI decoding apparatus 1300 according to an embodiment may include a receiving unit 1310 and an AI decoding unit 1330 . The AI decoding unit 1330 may include a first decoding unit 1332 , an AI setting unit 1333 , and an AI upscaling unit 1334 .

Although FIG. 13 shows that the receiver 1310 and the AI decoder 1330 are separate devices, they may be implemented through one processor. In this case, it may be implemented as a dedicated processor, or it may be implemented through a combination of an AP or a general-purpose processor such as a CPU or GPU and S/W. In addition, in the case of a dedicated processor, it may be implemented including a memory for implementing an embodiment of the present disclosure or a memory processing unit for using an external memory.

Also, the receiver 1310 and the AI decoder 1330 may be configured with one or more processors. In this case, it may be implemented as a combination of dedicated processors, or may be implemented through a combination of S/W with a plurality of general-purpose processors such as an AP, CPU, or GPU.

Similarly, the AI upscaling unit 1334, the AI setting unit 1333, and the first decoding unit 1332 may be implemented with different processors, respectively.

The receiver 1310 may correspond to the receiver 210 described above with reference to FIG. 2 . That is, the receiver 1310 may perform at least a part of the operation of the receiver 210 . However, the present invention is not limited thereto, and an operation different from the operation of the above-described receiver 210 may be performed.

The receiver 1310 receives a packetized video bitstream, obtains a video bitstream, and outputs the received video bitstream to the AI decoder 1330 . In this case, the packetized video bitstream may be packetized based on various network protocols such as Real-Time Transport Protocol (RTP).

Specifically, the receiver 1310 receives a packetized video bitstream through a network. The packetized video bitstream may include AI encoded data generated as a result of AI encoding. For example, the packetized video bitstream may include a main bitstream including image data and a sub-bitstream including AI data.

The main bitstream may be included in a predetermined main bitstream area in a video bitstream defined in a predetermined codec. The predetermined main bitstream region may be a space defined to include image data generated by generally encoding the original image 105 in a predetermined codec. For example, the predetermined main bitstream region includes a video parameter set (VPS), a sequence parameter set (SPS), a picture parameter set (PPS), and an I/P when the predetermined codec is an H.264, HEVC, and VVC codec. It may be a NAL unit area of the /B slice.

Meanwhile, the sub-bitstream region may be a space defined to include additional data excluding image data generated by encoding the original image 105 in a predetermined codec in general.

For example, when the predetermined codec is an H.264, HEVC, and VVC codec, the predetermined sub-bitstream area may be a Supplemental Enhancement Information (SEI) message area. In this case, the SEI message may include an SEI message including user data registered by a predetermined standard and an SEI message including unregistered user data identified by a Universally Unique Identifier (UUID).

For example, when the predetermined codec is the AV1 codec, the predetermined sub-stream may be metadata Open Bitstream Unit (OBU).

Meanwhile, a packet of a packetized video bitstream may include a single NAL unit packet, a fragmentation unit packet, or an aggregation packet. In this case, a plurality of fragmentation unit packets may include data of a single NAL unit, and one aggregation packet may include data of a plurality of NAL units.

The receiver 1310 may read a header of data and obtain a video bitstream from the data. For example, the receiver 1310 may include video bits included in a payload except for an Ethernet header, an IP header, a UDP/TCP header, an RTP header, etc. from the packetized video bitstream. stream can be obtained.

The video bitstream may include at least one NAL unit, and the NAL unit may include a Video Coded Layer (VCL) NAL unit and a non-VCL NAL unit. In this case, the VCL NAL unit means a part including video data encoded with respect to an actual video, and the non-VCL NAL unit is a part including metadata of the video, and is common to at least one frame such as VPS SPS, PPS, etc. It may include image parameter information used as , or supplemental data required for image decoding.

The receiver 1310 may output a video bitstream to the first decoder 1330 by arranging a plurality of NAL units in order. At this time, the receiving unit 1310 arranges the NAL units of a sequence parameter set (SPS), a picture parameter set (PPS), an I slice, a P slice, and a B slice in the order of the NAL unit to arrange the video bitstream in the first decoder 1332 . ) can be printed.

In this case, the image data included in the AI encoded data is image data generated through a predetermined codec (eg, MPEG-2, H.264, MPEG-4, HEVC, VC-1, VP8, VP9, or AV1). may be confirmed as In this case, the corresponding information may be transmitted to the first decoder 1332 through the receiver 1310 so that the image data can be processed by the identified codec.

In one embodiment, the data acquired by the receiving unit 1310 is a hard disk, a magnetic medium such as a floppy disk and a magnetic tape, an optical recording medium such as a CD-ROM and a DVD, and a magnetic-type disk such as a floppy disk. It may be obtained from a data storage medium including a magneto-optical medium or the like. For example, the receiver 1310 obtains a video bitstream including AI-encoded data from a video file having a file format such as mp4 or mov recorded on a data storage medium (eg, through demuxing). video bitstream).

The first decoding unit 1332 may correspond to the first decoding unit 234 described above with reference to FIG. 2 . The first decoding unit 1332 may further include the parsing unit 232 described above with reference to FIG. 2 . However, the present invention is not limited thereto, and a parsing unit (not shown) may exist separately from the first decoding unit 1332 . In this case, the parsing unit (not shown) may correspond to the above-described parsing unit 232 .

The first decoder 1332 may parse the video bitstream to obtain a sub-bitstream including AI data and a main bitstream including image data generated as a result of the first encoding of the first image 115 . . The first decoder 1332 may reconstruct the second image 135 corresponding to the first image 115 based on the image data. For example, the first decoder 1332 reconstructs the second image 135 corresponding to the first image 115 based on image data included in the main bitstream.

Meanwhile, the first decoding unit 1332 may obtain AI data from the sub-bitstream including the AI data, and may output the AI data to the AI setting unit 1333 . In this case, the AI data may include an AI upscaling activation flag and AI sub data.

The AI setting unit 1333 may correspond to the AI setting unit 238 described above with reference to FIG. 2 . The AI setting unit 1333 may determine whether to upscale and an upscale target based on the AI data.

When the AI setting unit 1333 determines whether to perform AI upscaling on the second image 135 based on the AI upscaling activation flag information, and determines to perform AI upscaling on the second image 135 , , controls the first decoder 1332 to provide the second image 135 to the AI upscaler 1334 .

When the AI decoder 1330 determines to perform AI upscaling on the second image 135, mode information (eg, prediction mode information, motion information, etc.) included in the image data, quantization parameter information, and AI data and the like may be provided from the first decoding unit 1332 to the AI setting unit 1333 .

Meanwhile, when the AI decoder 1330 determines not to perform AI upscaling on the second image 135 , the AI decoder 1330 outputs the second image 135 generated based on the image data. can be In this case, whether the second image 135 is provided to the AI upscaler 1334 or output from the AI decoder 1330 may be controlled through the AI setting unit 1333 .

The AI setting unit 1333 that has received the AI sub data provides DNN setting information and the like to the AI upscaling unit 1334 so as to AI upscale the second image 135 based on the AI sub data. Depending on the embodiment, the DNN setting information is obtained by further using the first decoding-related information such as mode information and quantization parameter information included in the image data, and the DNN setting information is provided to the AI upscaling unit 1334. can

The AI sub data provided to the AI setting unit 1333 includes information enabling AI upscaling of the second image 135 . In this case, the upscale target should correspond to the downscale target of the first DNN. Therefore, the AI sub data should include information that can confirm the downscale target of the first DNN.

The AI sub data may include information on a difference between the resolution of the original image 105 and the resolution of the first image 115 . Also, the AI sub data may include information related to the first image 115 . Also, the AI sub data may include resolution information of the third image 145 (or the original image 105 ) to be generated by the AI decoding apparatus 1300 .

The difference information may be expressed as information about a resolution conversion degree of the first image 115 compared to the original image 105 (eg, resolution conversion rate information), for example, the original image 105 compared to the first image 115 . It may be expressed as ratio (or multiple) information of one image 115 . That is, the ratio information of the first image 115 to the original image 105 may be flag information or index information indicating that the ratio of the original image 105 to the first image 115 is 1.5 times or 2 times. . And, since the resolution of the first image 115 is known through the resolution of the restored second image 135 and the degree of resolution conversion can be confirmed through this, the difference information can be expressed only with the resolution information of the original image 105 . may be Here, the resolution information may be expressed in horizontal and vertical sizes, or in a ratio (16:9, 4:3, etc.) and a size of one axis. In addition, when there is preset resolution information, it may be expressed in the form of an index or a flag.

The information related to the first image 115 is information on at least one of a bit rate of image data generated as a result of the first encoding of the first image 115 and a codec type used for the first encoding of the first image 115 . may include.

The AI setting unit 1333 selects an upscale target of the second image 135 based on at least one of difference information included in the AI data, information related to the first image 115 , and resolution information of the third image 145 . can decide The upscaling target may indicate, for example, to which resolution the second image 135 should be AI upscaled.

When the upscale target is determined, the AI setting unit 1333 AI upscales the second image 135 through the second DNN to generate a third image 145 corresponding to the upscale target. The DNN setting information is transmitted to the scale unit 1334 .

In an embodiment, the AI setting unit 1333 of the quantization parameter used for the first encoding of the first image included in the AI sub data to confirm the second DNN for the AI upscaling of the second image 135 . DNN setting information can be obtained by further inputting the average value.

In an embodiment, the AI setting unit 1333 further adds quantization parameter indicator information of at least a part of the first image included in the AI sub data to confirm the second DNN for AI upscaling of the second image 135 . You can obtain DNN configuration information by inputting it.

In an embodiment, the AI setting unit 1333 further adds information indicating the genre of the first image 115 included in the AI sub data to confirm the second DNN for AI upscaling of the second image 135 . You can obtain DNN configuration information by inputting it.

In an embodiment, the AI setting unit 1333 may obtain DNN setting information by further inputting additional metadata related to AI upscaling in order to confirm the second DNN for AI upscaling of the second image 135 . . The additional metadata may be further input for effective AI upscaling of the second image 135 .

13B is a diagram illustrating a configuration of an AI decoding apparatus 1350 according to another embodiment.

Referring to FIG. 13B , the AI decoding apparatus 1350 may include a configuration similar to that of FIG. 13A , except that a separate parsing unit 1381 is included.

The parsing unit 1381 parses the video file including the AI encoded data to obtain a metadata box including AI data and a media data box including image data from the video file, and extracts the image data included in the media data box. It may be transmitted to the first decryption unit 1382 , and AI data included in the metadata box may be transmitted to the AI setting unit 1383 . Other operations of the AI decoding apparatus 1350 are similar to those of the AI decoding apparatus 1300 , and thus detailed descriptions of the operations of the AI decoding apparatus 1350 will be omitted.

Hereinafter, the AI encoding apparatus 1400 for AI encoding of the original image 105 will be described with reference to FIG. 14 .

14 is a block diagram illustrating a configuration of an AI encoding apparatus 1400 according to an embodiment.

Referring to FIG. 14 , the AI encoding apparatus 1400 may include an AI encoding unit 1410 and a transmission unit 1430 . The AI encoder 1410 may include an AI downscaler 1412 , an AI setting unit 1413 , and a first encoder 1414 .

14 illustrates the AI encoder 1410 and the transmitter 1430 as separate devices, the AI encoder 1410 and the transmitter 1430 may be implemented through one processor. In this case, the AI encoder 1410 and the transmitter 1430 may be implemented as separate dedicated processors, or may be implemented through a combination of S/W with an AP, a general-purpose processor such as a CPU or GPU. In addition, in the case of a dedicated processor, it may be implemented including a memory for implementing an embodiment of the present invention, or may be implemented including a memory processing unit for using an external memory.

Also, the AI encoder 1410 and the transmitter 1430 may include one or more processors. In this case, it may be implemented as a combination of dedicated processors, or may be implemented through a combination of S/W with a plurality of general-purpose processors such as an AP, CPU, or GPU. Similarly, the AI downscaler 1412 and the first encoder 1414 may be implemented with different processors, respectively.

The AI encoder 1410 performs AI downscaling of the original image 105 and first encoding of the first image 115 , and includes a sub-bitstream including AI data and a main bitstream including image data. and transmits the video bitstream to the transmission unit 1430 .

The image data includes data generated as a result of the first encoding of the first image 115 . The image data may include data generated based on pixel values in the first image 115 , for example, residual data that is a difference between the first image 115 and prediction data of the first image 115 . . Also, the image data includes information used in the first encoding process of the first image 115 . For example, the image data includes mode information (eg, prediction mode information, motion information, etc.) used for first encoding the first image 115 and for first encoding the first image 115 . Information related to the quantization parameter used may be included. Also, the image data may include parameter information commonly used for at least one frame.

The AI data may include AI upscale activation flag information and AI sub data. The AI upscaling activation flag information may be flag information indicating whether AI upscaling is performed. That is, the AI data may include flag information indicating whether the first encoded image is an AI downscaled image.

The AI sub data includes information that enables the AI upscaling unit 1334 of the AI decoding apparatus 1300 to AI upscale the second image 135 to an upscale target corresponding to the downscale target of the first DNN. do. In an example, the AI sub data may include difference information between the original image 105 and the first image 115 . Also, the AI sub data may include information related to the first image 115 . The information related to the first image 115 is used for the resolution of the first image 115 , the bit rate of image data generated as a result of the first encoding of the first image 115 , and the first encoding of the first image 115 . information on at least one of the specified codec types may be included. Also, the AI sub data may include an average value of quantization parameters used for the first encoding of the first image. Also, the AI sub data may include quantization parameter indicator information of at least a part of the first image. Also, the AI sub data may include genre information of the first image.

Also, the AI data may include resolution information of the third image 145 (or the original image 105) to be generated through AI upscaling. Also, the AI data may further include additional metadata information for AI upscaling of the second image 135 .

The AI setting unit 1413 determines whether the AI downscale unit 1412 AI downscales the original image 105 through the first DNN.

The AI setting unit 1413 performs AI downscaling based on the resolution of the original image 105 , the type of the original image 105 (eg, the type of file), the type of the subject included in the original image 105 , and the like. can decide whether For example, when the resolution of the original image 105 is smaller than a predetermined resolution (eg, HD), the AI setting unit 1413 may determine not to perform AI downscaling.

When the AI setting unit 1413 determines to AI downscale the original image 105 through the first DNN, the AI downscaled first image 115 is generated from the original image 105 through the first DNN. The DNN configuration information may be provided to the AI downscale unit 1412 .

The AI setting unit 1413 further inputs the average value of the quantization parameter used for the first encoding of the first image to identify the first DNN for AI downscaling of the original image 105 to obtain DNN setting information, The DNN configuration information may be provided to the AI downscale unit 1412 .

The AI setting unit 1413 further inputs quantization parameter indicator information of at least a part of the first image to identify the first DNN for AI downscaling of the original image 105 to obtain DNN setting information, and DNN setting information may be provided to the AI downscale unit 1412 .

The AI setting unit 1413 further inputs information indicating the genre of the first image 115 to identify the first DNN for AI downscaling of the original image 105 to obtain DNN setting information, and DNN setting information may be provided to the AI downscale unit 1412 .

The AI setting unit 1413 obtains DNN setting information by further inputting additional metadata information to identify the first DNN for AI downscaling of the original image 105, and sets the DNN setting information to the AI downscaling unit 1412 ) can be provided.

The AI setting unit 1413 may determine a downscale target of the original image 105 based on a predetermined criterion.

In order to generate the first image 115 matching the downscale target, the AI setting unit 1413 may store a plurality of settable DNN setting information in the first DNN. The AI setting unit 1413 may obtain DNN setting information corresponding to a downscale target among a plurality of DNN setting information, and transmit the obtained DNN setting information to the AI downscale unit 1412 . The AI downscaling unit 1412 AI downscales the original image 105 through the first DNN set with the obtained DNN setting information.

Each of the plurality of pieces of DNN configuration information may be trained to generate the first image 115 having a predetermined resolution and/or a predetermined image quality. For example, the DNN setting information of any one of the plurality of DNN setting information is the first image 115 having a resolution that is 1/2 times smaller than the resolution of the original image 105, for example, 4K (4096*2160) may include information for generating the first image 115 of 2K (2048 * 1080) that is 1/2 times smaller than the original image 105 of A first image 115 of a resolution of 1/4 times smaller than that of a first image 115, for example, a first image 115 of 2K (2048*1080) which is 1/4 times smaller than an original image 105 of 8K (8192*4320). ) may include information for generating

The AI downscaling unit 1412 sets the first DNN as the DNN setting information obtained for the AI downscaling of the original image 105, and converts the first image 115 of a predetermined resolution and/or a predetermined quality to the first DNN. can be created through When DNN setting information for AI downscaling of the original image 105 is obtained among a plurality of DNN setting information, each layer in the first DNN may process input data based on information included in the DNN setting information. .

In one embodiment, in order to generate the first image 115 matching the downscale target, the AI downscaling unit 1412 is based on the DNN setting information obtained in the AI setting unit 1413, a plurality of pre-stored Among the first DNNs, a first DNN for AI downscaling the original video 105 may be determined, and the original video 105 may be AI downscaled through the determined first DNN. The plurality of first DNNs may have different layer structures or different preset parameters.

The AI setting unit 1413 is an original image (eg, a criterion determined based on at least one of compression ratio, compression quality, compression history information, and the type of the original image) based on a predetermined criterion) among the plurality of first DNNs. 105) may select a first DNN for downscaling the AI, and may transmit DNN configuration information related to the selected first DNN to the AI downscaling unit 1412 .

Hereinafter, a method for the AI setting unit 1413 to determine the downscale target will be described. The downscale target may indicate, for example, how much the resolution of the first image 115 is reduced from the original image 105 should be generated.

In an embodiment, the AI setting unit 1413 may determine a downscale target based on at least one of a compression rate, compression quality, compression history information, and a type of the original image 105 .

In one example, the AI setting unit 1413 may determine a downscale target based on a preset or a compression rate or compression quality input from a user.

As another example, the AI setting unit 1413 may determine the downscale target by using the compression history information stored in the AI encoding apparatus 1400 . For example, according to the compression history information available to the AI encoding apparatus 1400, a user's preferred encoding quality or compression rate may be determined, and a downscale target may be determined according to the encoding quality determined based on the compression history information. can For example, the resolution and quality of the first image 115 may be determined according to the encoding quality that has been most frequently used according to the compression history information.

As another example, the AI setting unit 1413 sets the encoding quality that has been used more than a predetermined threshold value (eg, average quality of encoding quality that has been used more than a predetermined threshold value) according to the compression history information. A downscale target may be determined based on it.

As another example, the AI setting unit 1413 may determine the downscale target based on the resolution and type (eg, file format) of the original image 105 .

In an embodiment, when the original image 105 is composed of a plurality of frames, the AI setting unit 1413 may independently determine a downscale target for each predetermined number of frames, or a common downscale target for all frames may decide.

In one example, the AI setting unit 1413 may classify frames constituting the original image 105 into a predetermined number of groups, and independently determine a downscale target for each group. The same or different downscale targets may be determined for each group. The number of frames included in the groups may be the same or different for each group.

In another example, the AI setting unit 1413 may independently determine a downscale target for each frame constituting the original image 105 . The same or different downscale target may be determined for each frame. The aforementioned AI downscale unit 1432 and AI setting unit 1433 may correspond to the AI downscale unit 712 and AI setting unit 718 .

Meanwhile, the first encoder 1414 may correspond to the first encoder 714 . The present invention is not limited thereto, and may further include a data processing unit 716 . However, the present invention is not limited thereto, and a data processing unit (not shown) may exist separately from the first encoding unit 1414 . In this case, the data processing unit (not shown) may correspond to the above-described data processing unit 716 .

The first encoder 1414 may obtain AI data from the AI setting unit 1313 . The first encoder 1414 may obtain image data by performing a first encoding of the first image 115 , and may transmit AI-encoded data including the image data and AI data to the transmitter 1430 . The first encoder 1414 may transmit a video bitstream including a sub bitstream including AI data and a main bitstream including image data to the transmitter 1430 .

The transmitter 1430 may generate a packetized bitstream by packetizing the video bitstream. The video bitstream includes at least one NAL unit, and in this case, the NAL unit of the parameter set commonly used for at least one frame, the NAL unit of the I slice, the P slice, and the NAL unit of the B slice may be received in order. .

The transmission unit 1430 includes at least one NAL unit included in the bitstream in the payload of the packet, and a header such as an Ethernet header, an IP header, a UDP/TCP header, an RTP header, etc. In addition, a packetized video bitstream may be generated, and the transmitter 1430 may transmit the packetized bitstream to the outside through a network.

14B is a diagram illustrating a configuration of an AI encoding apparatus 1450 according to another embodiment.

Referring to FIG. 14B , the AI encoding apparatus 1450 may include a configuration similar to that of FIG. 14A , except that a separate data processing unit 1465 is included.

The data processing unit 1465 may acquire AI data from the AI setting unit 1463 . The data processor 1465 may obtain image data from the first encoder 1464 . The data processing unit 1465 may generate a video file including AI data and image data. The data processing unit 1465 may transmit the generated video file to the transmission unit 1430 . In this case, the metadata box of the video file may include AI data, and the media data box of the video file may include image data. Other operations of the AI encoding apparatus 1450 are similar to the operations of the AI encoding apparatus 1400 , and thus a detailed description of the operation of the AI encoding apparatus 1450 will be omitted.

15A is a flowchart illustrating an AI decoding method according to an embodiment.

In operation S1505, the AI decoding apparatus 1300 may obtain a video bitstream including a main bitstream including image data obtained as a result of the first encoding of the first image and a sub-bitstream including AI data.

The main bitstream is included in a predetermined main bitstream area in a video bitstream defined for image encoding information in a predetermined codec, and the sub bitstream is a predetermined sub-bitstream in a video bitstream defined for additional data in a predetermined codec. It may be included in the bitstream area. In this case, the sub-bitstream may be a Supplemental Enhancement Information (SEI) message of the MPEG video codec. The SEI message may be an SEI message including user data registered by a predetermined standard or unregistered user data identified by a Universally Unique Identifier (UUID). Meanwhile, in the case of a codec such as AV1 or VP9, which is a predetermined codec, the sub-bitstream may be metadata Open Bitstream Unit (OBU). In this case, the metadata OBU may be a metadata OBU including user data registered according to a predetermined standard.

AI sub data included in the sub bitstream includes image genre information indicating the genre of the image, average quantization parameter information indicating the average quantization value of the first image, quantization parameter indicator information indicating the degree of quantization of at least a portion of the first image, and the original Image resolution information indicating the resolution of at least one of the image and the first image, image ratio information indicating the ratio of the original image and the first image, codec information indicating the codec used for encoding the first image, AI upscaling related At least one of metadata flag information indicating whether additional data is included and additional metadata information related to AI upscaling may be included. In this case, the quantization parameter indicator information may be index information indicating one quantization parameter group among a plurality of predetermined quantization parameter groups.

The quantization parameter indicator information may be index information indicating a quantization parameter group including a representative value of at least one quantization parameter among a subgroup in a frame, a frame, a frame group, and a video sequence.

In this case, the video sequence is a data unit including all consecutive frames of image data. A frame group is a data unit composed of one or more consecutive frames that share parameter information of a frame group parameter set. The frame group may be a group of pictures (GOP) or a coded video sequence (CVS). The frame group includes an coded instantaneous decoding refresh (IDR) frame or an intra random access pictures (IRAP) frame without reference to other frames. And the remaining frames of the frame group are coded with reference to the IDR frame (or IRAP frame). A frame represents one still image included in the video. A frame subgroup is a data unit composed of at least one block included in a frame.

In operation S1510, the AI decoding apparatus 1300 may first decode image data included in the main bitstream to obtain a second image corresponding to the first image.

In step S1515 , the AI decoding apparatus 1300 may acquire an AI upscaling activation flag included in AI data of the sub-bitstream.

In step S1520 , the AI decoding apparatus 1300 may determine whether the AI upscale activation flag information indicates activation of the AI upscale.

In step S1525, when the AI upscaling activation flag information indicates activation of AI upscaling, the AI decoding apparatus 1300 determines at least one of at least a portion of the image data and at least a portion of the AI sub data included in the sub bitstream. The AI upscaled third image may be acquired through the upscaling DNN set according to the selected upscaling DNN information among a plurality of upscaling DNN information stored in advance as .

The AI decoding apparatus 1300 provides quantization parameter information obtained from image data of the main bitstream and image genre information included in AI sub data, average quantization parameter information, quantization parameter indicator information, image resolution information, image ratio information, and AI up. Based on at least one of the scale-related additional metadata information, the upscaling DNN may be set as one upscaling DNN information among a plurality of upscaling DNN information.

The AI decoding apparatus 1300 may set the upscaling DNN as upscaling DNN information corresponding to one quantization group indicated by index information of quantization parameter indicator information among a plurality of upscaling DNN information.

In step S1530 , the AI decoding apparatus 1300 may output a third image.

In step S1535 , the AI decoding apparatus 1300 may output a second image when the AI upscaling activation flag information indicates that the AI upscaling is not activated.

15B is a flowchart illustrating an AI decoding method according to another embodiment.

In operation S1555, the AI decoding apparatus 1350 may obtain a video file including a media data box including image data obtained as a result of the first encoding of the first image and a metadata box including AI data.

In operation S1560, the AI decoding apparatus 1350 may first decode image data included in the media data box to obtain a second image corresponding to the first image.

In step S1565 , the AI decoding apparatus 1350 may acquire an AI upscaling activation flag included in the AI data of the metadata box.

In operation S1570, the AI decoding apparatus 1350 may determine whether the AI upscale activation flag information indicates activation of the AI upscale.

In step S1575, when the AI upscaling activation flag information indicates activation of AI upscaling, the AI decoding apparatus 1350 determines at least one of at least a portion of the image data and at least a portion of the AI sub data included in the metadata box. The AI upscaled third image may be acquired through the upscaling DNN set according to the selected upscaling DNN information among a plurality of upscaling DNN information stored in advance as .

In step S1580 , the AI decoding apparatus 1350 may output a second image when the AI upscaling activation flag information indicates that AI upscaling is not activated.

16A is a flowchart illustrating an AI encoding method according to an embodiment.

In step S1605, the AI encoding apparatus 1400 may determine whether to downscale the AI of the image.

In operation S1610 , the AI encoding apparatus 1400 may acquire the AI downscaled first image from the original image using the downscaling DNN based on the determination result.

The AI encoding apparatus 1400 includes quantization parameter information obtained from image data of the main bitstream and image genre information included in AI sub data, average quantization parameter information, quantization parameter indicator information, image resolution information, image ratio information, and AI up. A DNN for downscaling may be set as one of the downscaling DNN information among a plurality of downscaling DNN information based on at least one of the scale-related additional metadata information.

The AI encoding apparatus 1400 may set the DNN for downscaling as downscaling DNN information corresponding to one quantization group indicated by index information of quantization parameter indicator information among a plurality of downscaling DNN information.

In operation S1615, the AI encoding apparatus 1400 may obtain image data by encoding the first image.

In operation S1620, the AI encoding apparatus 1400 may generate an AI upscaling activation flag indicating whether AI upscaling corresponding to downscaling.

In step S1625 , when it is determined to perform AI downscaling on the original image, the AI encoding apparatus 1400 may generate AI sub data related to AI upscaling corresponding to the AI downscaling.

In operation S1630 , the AI encoding apparatus 1400 may generate AI data including an AI upscaling activation flag and AI sub data.

In operation S1635 , the AI encoding apparatus 1400 may generate a video bitstream including a main stream including image data and a sub bitstream including AI data. The main bitstream is included in a predetermined main bitstream area in a video bitstream defined for image encoding information in a predetermined codec, and the sub bitstream is a predetermined sub-bitstream in a video bitstream defined for additional data in a predetermined codec. It may be included in the bitstream area. In this case, the sub-bitstream may be an SEI message of an MPEG video codec. The SEI message may be an SEI message including user data registered by a predetermined standard or unregistered user data identified by a Universally Unique Identifier (UUID). Meanwhile, in the case of a codec such as AV1 or VP9, which is a predetermined codec, the sub-bitstream may be metadata Open Bitstream Unit (OBU). In this case, the metadata OBU may be a metadata OBU including user data registered according to a predetermined standard.

AI sub data included in the sub bitstream includes image genre information indicating the genre of the image, average quantization parameter information indicating the average quantization value of the first image, quantization parameter indicator information indicating the degree of quantization of at least a portion of the first image, and the original Image resolution information indicating the resolution of at least one of the image and the first image, image ratio information indicating the ratio of the original image and the first image, codec information indicating the codec used for encoding the first image, AI upscaling related At least one of metadata flag information indicating whether additional data is included and additional metadata information related to AI upscaling may be included. In this case, the quantization parameter indicator information may be index information indicating one quantization parameter group among a plurality of predetermined quantization parameter groups.

The quantization parameter indicator information may be index information indicating a quantization parameter group including a representative value of at least one quantization parameter among a subgroup in a frame, a frame, a frame group, and a video sequence.

16B is a flowchart illustrating an AI encoding method according to another embodiment.

In operation S1655 , the AI encoding apparatus 1450 may determine whether to downscale the image by AI.

In operation S1660, the AI encoding apparatus 1450 may acquire the AI downscaled first image from the original image using the downscaling DNN based on the determination result.

In operation S1665, the AI encoding apparatus 1450 may obtain image data by encoding the first image.

In operation S1670, the AI encoding apparatus 1450 may generate an AI upscaling activation flag indicating whether AI upscaling corresponding to downscaling.

In operation S1675, when it is determined to perform AI downscaling on the original image, the AI encoding apparatus 1450 may generate AI sub data related to AI upscaling corresponding to the AI downscaling.

In operation S1680 , the AI encoding apparatus 1450 may generate AI data including an AI upscaling activation flag and AI sub data.

In operation S1685, the AI encoding apparatus 1450 may generate a video file including a media data box including image data and a metadata box including AI data.

17 illustrates the corresponding relationship between the image data 1700 and the AI data 1740, along with the structures of the image data 1700 and the AI data 1740. In FIG.

A video-frame group-frame hierarchical structure of the image data 1700 is described in FIG. 17 . The video 1702 of FIG. 17 is a data unit including all consecutive frames of the image data 1700 . Parameter information of the video parameter set may be applied to all frames included in the video 1702 . The video parameter set is included in the video header 1704 .

Video 1702 may include a plurality of groups of frames. A frame group is a data unit composed of one or more consecutive frames that share parameter information of a frame group parameter set. The frame group may be a group pf pictures (GOP) or a coded video sequence (CVS). The frame group parameter set may be included in the frame group header. For example, the frame group parameter set of the first frame group 1710 may be included in the first frame group header 1712 . In addition, the frame group parameter set of the second frame group 1714 may be included in the second frame group header 1716 .

One frame group is independently encoded without reference to another frame group. The frame group includes an coded instantaneous decoding refresh (IDR) frame or an intra random access pictures (IRAP) frame without reference to other frames. And the remaining frames of the frame group are coded with reference to the IDR frame (or IRAP frame). Accordingly, the first frame group 1710 is independently coded without reference to other frame groups of the video 1702 . And the first frame 1720, which is the first encoded frame of the first frame group 1710, is an IDR frame (or IRAP frame). The remaining frames of the first frame group 1710 including the second frame 1730 are encoded with reference to the first frame 1720 .

A frame represents one still image included in the video. The frame header may include a frame parameter set including parameter information applied to the frame. For example, the first frame header 1722 of the first frame 1720 may include a frame parameter set applied to the first frame 1720 . Similarly, the second frame header 1732 of the second frame 1730 may include a frame parameter set applied to the second frame 1730 .

A method of classifying the AI data 1740 according to the video-frame group-frame hierarchy described above in FIG. 17 is described. AI data 1740 may be classified into video AI data 1742 , frame group AI data 1750 , and frame AI data 1760 according to an application range. The video AI data 1742 refers to AI data commonly applied to all frame groups included in a video. In addition, the frame group AI data 1750 refers to AI data commonly applied to frames included in the current frame group. Also, the frame AI data 1760 refers to AI data applied to the current frame.

Video AI data 1742 corresponds to video header 1704 . Accordingly, the video AI data 1742 may be decoded in parallel with the video header 1704 . Alternatively, the video AI data 1742 may be decoded immediately before decoding of the video header 1704 . Alternatively, the video AI data 1742 may be decoded immediately after decoding of the video header 1704 .

The frame group AI data 1750 corresponds to the frame group header. Referring to FIG. 17 , the first frame group AI data 1752 corresponds to the first frame group header 1712 . And the second frame group AI data 1754 corresponds to the second frame group header 1716 . The first frame group AI data 1752 and the second frame group AI data 1754 may be decoded in parallel with the first frame group header 1712 and the second frame group header 1716, respectively. Alternatively, the first frame group AI data 1752 and the second frame group AI data 1754 may be decoded immediately before decoding of the first frame group header 1712 and the second frame group header 1716 , respectively. Alternatively, the first frame group AI data 1752 and the second frame group AI data 1754 may be decoded immediately after decoding of the first frame group header 1712 and the second frame group header 1716 , respectively.

The frame AI data 1760 corresponds to the frame header. Referring to FIG. 17 , the first frame AI data 1762 corresponds to the first frame header 1722 . And the second frame AI data 1764 corresponds to the second frame header 1732 . The first frame AI data 1762 and the second frame AI data 1764 may be decoded in parallel with the first frame header 1722 and the second frame header 1732 , respectively. Alternatively, the first frame AI data 1762 and the second frame AI data 1764 may be decoded immediately before decoding of the first frame header 1722 and the second frame header 1732 , respectively. Alternatively, the first frame AI data 1762 and the second frame AI data 1764 may be decoded immediately after decoding of the first frame header 1722 and the second frame header 1732 , respectively.

According to an embodiment, the data processing unit 716 of FIG. 7 may generate AI encoded data in the form of a single file including both the image data 1700 and the AI data 1740 . In this specification, a file means a collection of data stored in a memory. And a video file means a collection of image data stored in a memory, and the image data may be implemented in the form of a bitstream.

According to an embodiment, the AI data 1740 may not be inserted into the image data 1700 and may be configured separately from the image data 1700 within a single file. Therefore, even though the AI-encoded data is composed of a single file, since the AI data 1740 and the image data 1700 are separated, the AI data 1740 and/or the image data 1700 are the AI data 1740 and the image data 1700. ), but is not limited thereto.

18 illustrates an embodiment of a video bitstream including AI data and image data.

The video bitstream 1800 includes a sub bitstream 1810 and a main bitstream 1830 . The sub bitstream 1810 includes additional information about image data included in the main bitstream 1830 .

The sub bitstream 1810 may include synchronization data 1815 and AI data 1820 .

The first decoder 1332 may extract the synchronization data 1815 and the AI data 1820 from the sub-bitstream 1810 . In addition, the first decoder 1332 may extract the image data 1831 from the main bitstream 1830 . The first decoding unit 1332 transmits the synchronization data 1815 to the AI upscaling unit 1334 or to the AI upscaling unit 1334 through the AI setting unit 1333, and the AI upscaling unit 1334 A, according to the synchronization data 1815 , the image data 1831 and the AI data 1820 may be synchronized.

AI data 1820 includes video AI data 1822 , frame group AI data 1824 , and frame AI data 1826 . The video AI data 1822 is set to correspond to the video header 1832 , the frame group AI data 1824 to the frame group header 1836 , and the frame AI data 1826 to the frame header 1840 .

According to an embodiment, when the same upscale DNN information is used for all frames of a video, the frame group AI data 1824 and the frame AI data 1826 may be omitted from the AI data 1820 . Alternatively, when the upscale DNN information is adaptively obtained for each frame group and the same upscale DNN information is used for all frames of the frame group, the frame AI data 1826 from the AI data 1820 may be omitted.

The synchronization data 1815 relates to the synchronization of the video AI data 1822, the frame group AI data 1824, and the frame AI data 1826 with the video header 1832, the frame group header 1836, and the frame header 1840. include information. For example, the synchronization data 1815 may include reproduction order information or decoding order information of the image data 1831 of the main bitstream 1830 . Accordingly, the AI setting unit 1333 may acquire upscale DNN information required for AI upscaling of the low-resolution image from the AI data determined according to the reproduction order information or the decoding order information of the synchronization data 1815 .

For example, in order to AI upscale the low-resolution image of the frame 1838 , the AI setting unit 1333 may set the frame group AI data 1824 and the frame corresponding to the frame group 1834 based on the synchronization data 1815 . Frame AI data 1826 corresponding to 1838 may be determined. And in consideration of the video AI data 1822 applied to all frames along with the frame group AI data 1824 and the frame AI data 1826 , the AI setting unit 1333 AI upscales the low-resolution image of the frame 1838 . It is possible to obtain upscale DNN information for

The AI data 1820 of the sub bitstream 1810 may be decoded before the image data 1831 of the main bitstream 1830 . Accordingly, the AI setting unit 1333 may acquire upscale DNN information before decoding the image data 1831 according to the AI data 1820 . The upscale DNN information may be applied to the entire video.

Alternatively, for adaptive AI upscaling in units of frame groups, different upscale DNN information may be obtained for each frame group. Alternatively, for frame-by-frame adaptive AI upscaling, upscale DNN information may be obtained differently for each frame in advance.

The AI setting unit 1333 may decode the AI data 1820 of the sub bitstream 1810 according to the decoding order of the image data 1831 of the main bitstream 1830 .

Decoding of the video AI data 1822 may be performed immediately before or after decoding of the video header 1832 . Alternatively, the video AI data 1822 may be decoded in parallel with the video header 1832 . In order to decode the video AI data 1822 according to the decoding order of the video header 1832 , the synchronization data 1815 may be referred to.

Similarly, the AI setting unit 1333 may perform decoding of the frame group AI data 1824 immediately before or after decoding of the frame group header 1836 according to the first decoding unit 1332 . Alternatively, the AI setting unit 1333 may decode the frame group AI data 1824 in parallel with the decoding of the frame group header 1836 according to the first decoding unit 1332 . In order to decode the frame group AI data 1824 according to the decoding order of the frame group header 1836, the synchronization data 1815 may be referred to.

In addition, the AI setting unit 1333 may perform decoding of the frame AI data 1826 immediately before or after decoding of the frame header 1840 according to the first decoding unit 1332 . Alternatively, the frame AI data 1826 may be decoded in parallel with the decoding of the frame header 1840 according to the first decoding unit 1232 . In order to decode the frame AI data 1826 according to the decoding order of the frame header 1840 , the synchronization data 1815 may be referred to.

19 shows an embodiment of AI encoded data when AI data and image data are separated in a single file. In FIG. 19 , AI-encoded data is included in a video file 1900 of a predetermined container format. The predetermined container format may be MP4, AVI, MKV, FLV, or the like.

The video file 1900 is composed of a metadata box 1910 and a media data box 1930 . The metadata box 1910 includes information about media data included in the media data box 1930 . For example, the metadata box 1910 may include information on a type of media data, a type of a codec used for encoding the media data, a media playback time, and the like.

The metadata box 1910 may include synchronization data 1915 and AI data 1920 . The synchronization data 1915 and the AI data 1920 are encoded according to an encoding method provided by a predetermined container format and stored in the metadata box 1910 .

The parsing unit 1381 may extract the synchronization data 1915 and the AI data 1920 from the metadata box 1910 . In addition, the parsing unit 1381 may extract the image data 1931 from the media data box 1930 . The parsing unit 1381 may transmit the image data 1931 to the first decoder 1332 according to the synchronization data 1915 , and transmit the AI data 1920 to the AI setting unit 1333 . Alternatively, the parsing unit 1381 transmits the synchronization data 1915 to the AI upscaling unit 1334 through the AI upscaling unit 1334 or the AI setting unit 1333, and the AI upscaling unit 1334 synchronizes According to the data 1915 , the image data 1931 and the AI data 1920 may be synchronized.

AI data 1920 includes video AI data 1922 , frame group AI data 1924 , and frame AI data 1926 . The video AI data 1922 is set to correspond to the video header 1932 , the frame group AI data 1924 to the frame group header 1936 , and the frame AI data 1926 to the frame header 1940 .

According to an embodiment, when the same upscale DNN information is used for all frames of a video, the frame group AI data 1924 and the frame AI data 1926 may be omitted from the AI data 1920 . Alternatively, when the upscale DNN information is adaptively obtained for each frame group unit and the same upscale DNN information is used for all frames of the frame group, the frame AI data 1926 may be omitted from the AI data 1920 .

The synchronization data 1915 relates to the synchronization of the video AI data 1922, the frame group AI data 1924, and the frame AI data 1926 with the video header 1932, the frame group header 1936, and the frame header 1940. include information. For example, the synchronization data 1915 may include playback order information or decoding order information of the image data 1931 of the media data box 1930 . Accordingly, the AI setting unit 1333 may acquire the upscale DNN information required for AI upscaling of the low-resolution image from the AI data determined according to the playback order information or the decoding order information of the synchronization data 1915, and the AI upscaling It may pass to some 1334 .

For example, in order to AI upscale the low-resolution image of the frame 1938, the parsing unit 1381 or the AI setting unit 1281 sets the frame group AI corresponding to the frame group 1934 based on the synchronization data 1915. Frame AI data 1926 corresponding to the data 1924 and the frame 1938 may be determined. And in consideration of the video AI data 1922 applied to all frames together with the frame group AI data 1924 and the frame AI data 1926, the AI setting unit 1281 AI upscales the low-resolution image of the frame 1938. It is possible to obtain upscale DNN information for

The AI data 1920 of the metadata box 1910 may be decoded earlier than the image data 1931 of the media data box 1930 . Accordingly, the AI setting unit 1383 may acquire upscale DNN information prior to decoding the image data 1931 according to the AI data 1920 . The upscale DNN information may be applied to the entire video. Alternatively, for adaptive AI upscaling in units of frame groups, different upscale DNN information may be obtained for each frame group. Alternatively, for frame-by-frame adaptive AI upscaling, upscale DNN information may be obtained differently for each frame in advance.

The AI setting unit 1383 may decode the AI data 1920 of the metadata box 1910 according to the decoding order of the image data 1931 of the media data box 1930 . Decoding of the video AI data 1922 may be performed immediately before or after decoding of the video header 1932 . Alternatively, the video AI data 1922 may be decoded in parallel with the video header 1932 . In order to decode the video AI data 1922 according to the decoding order of the video header 1932 , the synchronization data 1915 may be referred to.

Similarly, the AI setting unit 1383 may perform decoding of the frame group AI data 1924 immediately before or after decoding of the frame group header 1936 according to the first decoding unit 1382 . Alternatively, the AI setting unit 1383 may decode the frame group AI data 1924 in parallel with the decoding of the frame group header 1936 according to the first decoding unit 1382 . In order to decode the frame group AI data 1924 according to the decoding order of the frame group header 1936, the synchronization data 1915 may be referred to.

In addition, the AI setting unit 1383 may perform decoding of the frame AI data 1926 immediately before or after decoding of the frame header 1940 according to the first decoding unit 1382 . Alternatively, the AI setting unit 1383 may decode the frame AI data 1926 in parallel with the decoding of the frame header 1940 according to the first decoding unit 1382 . In order to decode the frame AI data 1926 according to the decoding order of the frame header 1940 , the synchronization data 1915 may be referred to.

In FIG. 19 , for convenience of description, it has been described that one metadata box 1910 and one media data box 1930 are included in the video file 1900 . However, the video file 1900 may include two or more metadata boxes and two or more media data boxes. Accordingly, two or more image data segments in which image data is divided into predetermined time units may be stored in two or more media data boxes. In addition, information on the image data segments stored in the two or more media data boxes may be included in the two or more meta data boxes. Also, two or more metadata boxes may each contain AI data.

20A illustrates an example of AI encoded data when AI data is inserted into image data.

The video file 2000, like the video file 1900 of FIG. 19 , includes a metadata box 1902 and a media data box 1904 . Since AI data is inserted into the image data, the metadata box 1902 does not include AI data. Instead, image data into which AI data is inserted is included in the media data box 1904 . AI data may be encoded according to a video codec of image data. For example, the video codec of the image data may be H.264, HEVC, AVS2.0, Xvid, or the like.

Accordingly, the parsing unit 232 does not extract AI data from the metadata box 2002 . Instead, the first decoder 234 may extract AI data from the image data 2005 and transmit the extracted AI data to the AI setting unit 238 . And, based on the upscale DNN information obtained by the AI data, the AI upscaling unit 236 may AI upscale the low-resolution image restored by the first decoding unit 234 .

Unlike the video file 1900 of FIG. 19 , since AI data is inserted in the image data 2005 according to the decoding order of the image data 2005 , the video file 2000 may not include synchronization data. Accordingly, the AI data is sequentially decoded as the image data 2005 is decoded.

The video AI data 2008 may be located immediately after the video header 2006 including parameter information of the video. Accordingly, the video AI data 2008 may be decoded after video parameters included in the video header 2006 are decoded. According to an embodiment, different from FIG. 16A , the video AI data 2008 may be located immediately before the video header 2006 .

The frame group AI data 2014 may be located immediately after the frame group header 2012 including parameter information of the frame group 2010 . Accordingly, the frame group AI data 2014 may be decoded after the frame group parameters included in the frame group header 2012 are decoded. According to an embodiment, unlike FIG. 20A , the frame group AI data 2014 may be located immediately before the frame group header 2012 . The decoding order of the frame group header and frame group AI data of the remaining frame groups decoded after the frame group 2010 may be determined in the same manner as the decoding order of the frame group header 2012 and the frame group AI data 2014 .

The frame AI data 2020 may be located immediately after the frame header 2018 including the parameter information of the frame 2016 . Accordingly, the frame AI data 2020 may be decoded after the frame group parameters included in the frame header 2018 are decoded. According to an embodiment, different from FIG. 20A , the frame AI data 2020 may be located immediately before the frame header 2018 . The decoding order of the frame header and frame AI data of the remaining frames decoded after the frame 2016 may be determined in the same manner as the decoding order of the frame header 2018 and the frame AI data 2020 .

20B illustrates another embodiment of AI encoded data when AI data is inserted into image data.

Like the video file 2000 of FIG. 20A , the video file 2020 includes a metadata box 2022 and a media data box 2024 . In addition, the metadata box 2022 does not include AI data, instead, image data 2025 into which AI data is inserted is included in the media data box 2024 . However, unlike the video file 2000 of FIG. 15A , AI data is inserted into the corresponding data header in the video file 2020 .

The video AI data 2028 may be included in the video header 2026 including parameter information of the video. Accordingly, the video AI data 2028 may be decoded together with the video parameters included in the video header 2026 .

The frame group AI data 2034 may be included in the frame group header 2032 including parameter information of the frame group 2030 . Accordingly, the frame group AI data 2034 may be decoded together with the frame group parameters included in the frame group header 2032 . Frame group AI data of the remaining frame groups decoded after the frame group 2030 may also be included in the frame group header.

The frame AI data 2040 may be included in a frame header 2038 including parameter information of the frame 2036 . Accordingly, the frame AI data 2040 may be decoded together with the frame group parameters included in the frame header 2038 .

Frame AI data of the remaining frames decoded after the frame 2036 may also be included in the frame header.

20C illustrates an embodiment of AI-encoded data when some AI data is inserted into image data and the remaining AI data is separated from image data.

Like the video file 2000 of FIG. 20A , the video file 2050 includes a metadata box 2052 and a media data box 2056 . Metadata box 2052 contains video AI data 2054 that applies to every frame of the video. In addition, the frame group AI data and the frame AI data are included in the image data of the media data box 2056 .

Unlike the video AI data 2008 of FIG. 20A and the video AI data 2028 of FIG. 20B , the video AI data 2054 included in the metadata box 2052 may be decoded prior to decoding of the image data. In addition, the frame group AI data and the frame AI data are sequentially decoded as the image data is decoded.

Accordingly, the parsing unit 232 may extract the video AI data 2054 from the metadata box 2052 . In addition, the parsing unit 232 may transmit the video AI data 2054 to the AI setting unit 238 . Also, the parser 232 may transmit the image data 2057 to the first decoder 234 . In addition, the first decoder 234 may reconstruct a low-resolution image by decoding the image data 2057 , and extract the frame group AI data 2064 and the frame AI data 2070 . Also, the first decoding unit 234 may transmit the frame group AI data 2064 and the frame AI data 2070 to the AI setting unit 238 . In addition, the AI setting unit 238 may acquire upscale DNN information for AI upscaling of a low-resolution image according to the video AI data 2054 , the frame group AI data 2064 , and the frame AI data 2070 .

The frame group AI data 2064 may be located immediately after the frame group header 2062 including parameter information of the frame group 2060 . However, according to an embodiment, the frame group AI data 2064 may be located immediately before the frame group header 2062 . Also, the frame group AI data 2064 may be included in the frame group header 2062 .

The frame AI data 2070 may be located immediately after the frame header 2068 including parameter information of the frame 2066 . However, according to an embodiment, the frame AI data 2070 may be located immediately before the frame header 2068 . Also, the frame AI data 2070 may be included in the frame header 2068 .

In FIG. 20C , it is expressed that only the video AI data 2054 is included in the metadata box 2052 , but frame group AI data may be additionally included in the metadata box 2052 . Alternatively, a part of the frame group AI data may be included in the metadata box 2052 . Alternatively, a part of the frame AI data may be included in the metadata box 2052 .

The frame group AI data 2014 and 2064 and the frame AI data 2020 and 2070 inserted into the media data box in FIGS. 20A and 20C may be inserted in the form of a Supplementary Enhancement Information (SEI) message. The SEI message is a data unit including additional information about an image that is not essential for decoding of the image. The SEI message may be transmitted in units of frame groups or frames. And the SEI message may be extracted from the parsing unit 232 and delivered to the AI setting unit 238 .

21 illustrates an embodiment of AI encoded data divided into video segment units when AI data and image data are separated as shown in FIG. 19 .

When the size of metadata including AI data is large or when video data is live streamed through a communication channel, in order to reduce the overhead required for processing metadata including AI data, AI-encoded data is divided into video segments. It may be stored in the video file 2100 . A video segment is a portion of a video and includes frames for a predetermined period of time. For example, a video segment may contain only one or more groups of frames. If the video segment includes one frame group, the video file 2100 may include as many video segments as the number of frame groups of image data.

Each video segment may include a metadata box and a media data box. Metadata including AI data is divided and stored in a plurality of video segments. Accordingly, AI data is divided and stored in metadata boxes for video segments as well as the metadata box 2110 for the entire image data. Therefore, when AI upscaling a low-resolution image of a specific part of image data, AI data includes the metadata box 2110 as compared to a case where all AI data is stored in the metadata box 2110 for the entire image data. When distributed and stored in metadata boxes for video segments, AI data for obtaining upscale DNN information suitable for the specific part may be obtained faster.

For example, when playing from the current video segment, for AI upscaling of the low-resolution image of the current video segment, the video AI data 2112 of the metadata box 2110 and the segment of the current video segment data box 2120 Only AI data in the metadata box 2130 is referenced. And AI data of metadata boxes of other video segments are not referenced. Accordingly, overhead according to the decoding of AI data may be reduced.

Also, when video data is live-streamed, AI data may not be transmitted all at once during initial playback, but may be transmitted in a distributed manner in units of video segments. Accordingly, since the AI data is distributed and transmitted sequentially, the overhead according to the decoding of the AI data can be reduced. Therefore, it may be advantageous to divide and transmit the video segment unit.

In FIG. 21 , the metadata box 2110 for the entire image data includes video AI data 2112 . Video AI data 2112 is applied to all video segments included in the video. The metadata box 2110 may be decoded earlier than the current video segment data box 2120 .

The current video segment data box 2120 includes a segment metadata box 2130 and a segment media data box 2140 . The segment metadata box 2130 may include synchronization data 2131 and AI data 2132 . And the segment media data box 2140 includes video segment data 2141 .

AI data 2132 of current video segment data box 2120 may include video segment AI data 2134 , frame group AI data 2136 , and frame AI data 2138 . The video segment AI data 2134 is applied to all frame groups included in the current video segment. And the frame group AI data 2136 is applied to all frames included in the current frame group. Frame AI data 2138 is applied to the current frame.

According to an embodiment, when the same upscale DNN information is used for all frames of a video segment, the frame group AI data 2136 and the frame AI data 2138 in the AI data 2132 of the current video segment data box 2120 ) can be omitted. Alternatively, when upscale DNN information is adaptively determined for each frame group and the same upscale DNN information is used for all frames of a frame group, frame AI data from AI data 2132 of the current video segment data box 2120 (2138) may be omitted.

According to an embodiment, when the video segment includes only one frame group, the video segment AI data 2134 may be omitted from the AI data 2132 of the current video segment data box 2120 . In addition, the frame group AI data 2136 may serve as the video segment AI data 2134 . Alternatively, when the video segment includes only one frame group and the same upscale DNN information is used for all frames of the frame group, the video segment AI data 2134 in the AI data 2132 of the current video segment data box 2120 and frame AI data 2138 may be omitted. And based on the upscale DNN information according to the frame group AI data 2136, AI upscaling may be applied to all frames of the video segment. The synchronization data 2131 includes the video segment AI data 2134 , the frame group AI data 2136 , and the frame AI data 2138 and the video segment header 2142 , the frame group header 2146 , and the frame header 2148 . includes information about The video segment header 2142 includes video segment parameters commonly applied to frames included in the video segment. For example, the synchronization data 2131 may include playback order information or decoding order information of the video segment data 2141 of the segment media data box 2140 . Accordingly, the AI setting unit 238 may acquire upscale DNN information required for AI upscaling of a low-resolution image from the AI data determined according to the synchronization data 2131 .

For example, in order to AI upscale the low-resolution image of the frame 2149 , the parsing unit 232 or the AI setting unit 238 may set the frame group AI corresponding to the frame group 2144 based on the synchronization data 2131 . The data 2136 and the frame AI data 2136 corresponding to the frame 2149 may be acquired. And in consideration of the video segment AI data 2134 applied to all frames together with the frame group AI data 2136 and the frame AI data 2138 , the AI setting unit 238 raises the AI of the low-resolution image of the frame 2149 . Upscale DNN information for scale can be obtained.

According to an embodiment, the AI data 2132 of the segment metadata box 2130 may be decoded earlier than the video segment data 2141 of the segment media data box 2140 . Accordingly, the AI setting unit 238 may acquire upscale DNN information before decoding the video segment data 2141 according to the AI data 2132 . The obtained upscale DNN information may be applied to the entire video segment. Alternatively, for adaptive AI upscaling in units of frame groups, upscale DNN information may be differently obtained in advance for each frame group. Alternatively, for frame-by-frame adaptive AI upscaling, upscale DNN information may be obtained differently for each frame in advance.

According to an embodiment, the AI setting unit 238 decodes the AI data 2132 of the segment metadata box 2130 according to the decoding order of the video segment data 2141 of the segment media data box 2140. can For example, the AI setting unit 238 may decode the frame group AI data 2136 according to the decoding order of the frame group header 2146 according to the first decoding unit 234 . The AI setting unit 238 may decode the frame AI data 2138 according to the decoding order of the frame header 2148 according to the first decoding unit 234 . With reference to the synchronization data 2131, the decoding order of the video segment AI data 2134, the frame group AI data 2136, and the frame AI data 2138 and the video segment header 2142, the frame group header 2146, the frame The decoding order of the header 2148 may be synchronized.

The remaining video segment data boxes after the current video segment data box 2120 may also be sequentially decoded in the same manner as the current video segment data box 2120 .

22 illustrates an embodiment of AI data 2240 and image data 200 transmitted as two separate files.

When the AI data 2240 is not received by the receiver 210 , the low resolution image obtained from the image data 2200 is not AI upscaled. If the AI data 2240 is received by the receiving unit 210 , upscale DNN information necessary for AI upscaling of the low resolution image is obtained according to the AI data 220 transmitted to the AI setting unit 238 .

The image data 2200 may include a video header 2210 , a frame group header 2222 of a frame group 2220 , and a frame header 2232 of a frame 2230 . The AI data 2240 may include video AI data 2242 , frame group AI data 2250 , and frame AI data 2260 . Since the image data 2200 and the AI data 2240 are transmitted as separate files, the image data 2200 and/or the AI data 2240 are synchronization data required for the synchronization of the image data 2200 and the AI data 2240 . may include The synchronization data may indicate a decoding order or a reproduction order of the image data 2200 .

The parsing unit 232 or the AI setting unit 238 sets the playback order or decoding order of the video AI data 2242, the frame group AI data 2250, and the frame AI data 2260 according to the synchronization data to the video header 2210. , may be set to match the playback order or decoding order of the frame group header 2222 and the frame header 2232 . A dotted line between the video AI data 2242 and the video header 2210 of FIG. 22 , a dotted line between the frame group AI data 2250 and the frame group header 2222 , and a dotted line between the frame AI data 2260 and the frame header 2232 . represents the synchronization between AI data and data header.

The image data 2200 and the AI data 2240 may include an identification number for matching the two data. For example, the AI data 2240 may include an identification number of the image data 2200 to which the AI data 2240 is applied. Conversely, the image data 2200 may include an identification number of the AI data 2240 applied to the image data 2200 . Alternatively, the identification number may be included in both the image data 2200 and the AI data 2240 . Accordingly, even when the image data 2200 and the AI data 2240 are not transmitted at the same time, the image data 2200 and the AI data 2240 may be matched with each other according to the identification number.

According to an embodiment, when upscale DNN information is acquired in units of frame groups, the frame AI data 2260 may be omitted from the AI data 2240 . In addition, when the same update DNN information is used for all frames of the video, the frame group AI data 2250 and the frame AI data 2260 may be omitted from the AI data 2240 .

23 is a diagram illustrating a structure of an SEI message according to an embodiment.

Referring to FIG. 23 , first, the SEI message may include itu_t_t35_country_code 2305 . The itu_t_t35_country_code 2305 may mean a byte specified as a country code by Annex A of the ITU-T T.35 standard. For example, the value may be 0xB5, which may be the US country code specified by Annex A of the ITU-T T.35 standard.

Next, the SEI message may include itu_t_t35_terminal_provider_code 2310 . The itu_t_t35_terminal_provider_code 2310 may be a 16-bit (2btye) field that specifies a code of a terminal provider or entity specified by the ITU-T T.35 standard. For example, the value may be 0x003C, which may be a code representing S company.

Next, the SEI message may include itu_t_t35_terminal_provider_oriented_code 2315 . itu_t_t35_terminal_provider_oriented_code means a code defined by a terminal provider or entity specified by the ITU-T T.35 standard. For example, the value may be 0x0002, which may indicate an AI codec (Codec) defined by S company.

Next, the SEI message may include application_identifier 2320 . application_identifier may be an identifier indicating one application method among various application methods of ST-2094, which is a standard for defining dynamic metadata of high dynamic range (HDR). In this case, the value may be 0x0004, which may indicate the S company application method of the ST 2094-40 standard.

Next, the SEI message may include AI data in the form of a predetermined syntax element. For example, the SEI message may include application_version 2325 . application_version 2325 may mean a genre of an image. For example, when the value is 0x0001, it may indicate that the genre of the image is a general image.

Next, the SEI message may include aid_flag 2330 . aid_flag 2330 may indicate whether AI downscaling is applied to an image and whether AI upscaling is performed. For example, the value may be 0X0001, which may indicate that AI downscaling has been applied to the image, and thus performing AI upscaling.

Next, the SEI message may include aid_average_qp 2335 . aid_average_qp 2335 is the average of the quantization parameters applied to the image (that is, applied during the first encoding of the original image), and the value may be 0x1F, which may indicate that the average of the quantization parameters applied to the image is 31.

Next, the SEI message may include aid_qp_selection 2340 . aid_qp_selection 2340 is indicator information indicating a quantization degree (or level) of at least a part of an image, and may be information in the form of an index. For example, the value may be 0x0000. A description of aid_qp_selection 2340 will be described later with reference to FIG. 21 .

Next, the SEI message may include aid_multiplication 2345 . aid_multiplication 2345 is information indicating the ratio (or multiple) between the original image and the first image, and may be information in the form of a flag or an index indicating one of the ratios (or multiples) between a plurality of original images and the first image. For example, the value may be 0x0000, and may indicate that the ratio between the original image and the first image is 2x.

Next, the SEI message may include aid_codec_information 2350 . aid_codec_information 2350 is information indicating a predetermined codec used for the first encoding of an image, and may be information in the form of an index. For example, the value may be 0x0000, which may indicate the H.264 codec.

Next, the SEI message may include aid_resolution_information 2355 . aid_resolution_information 2355 is information indicating the resolution of the original image (or the first image), and may be in various forms. For example, information in the form of an index indicating one of the resolutions of the plurality of original images (or the first image) can be For example, the value may be 0x0000, and may indicate that the resolution of the first image is Full HD.

Next, the SEI message may include aid_meta_flag 2360 . aid_meta_flag 2360 may mean a flag indicating whether additional metadata related to AI upscaling is included. For example, the value may be 0x0000, which may indicate that no additional metadata related to AI upscaling is included.

According to an embodiment, some of the AI upscaling related additional metadata may be located before the aid_meta_flag 2360 , and in this case, it may not be treated as additional metadata any longer. Alternatively, according to an embodiment, AI upscaling related data located before aid_meta_flag 2360 may be located after aid_meta_flag 2360 , and in this case, the corresponding data may be treated as additional metadata.

Previously, various data included in the SEI message have been described, but this is only an example, and those skilled in the art can understand that the SEI message may include some of them or may further include other data. It will be understood by those skilled in the art that the order may be changed from one another as long as there is no contradiction.

However, the present invention is not limited thereto, and a data structure applicable to the aforementioned AI data may be similar to that of the aforementioned SEI message.

24 shows a syntax structure table including AI data, according to an embodiment.

Referring to FIG. 24 , according to the aid_main syntax table 2400, syntax elements such as aid_flag, aid_average_qp, aid_qp_selection, aid_multiplication, aid_codec_information, aid_resolution_information, aid_meta_flag, etc. may be parsed.

aid_flag is an AI upscaling activation flag corresponding to aid_flag of FIG. 17 . When aid_flag indicates activation of AI upscaling (if(aid_flag)), syntax elements corresponding to AI sub-data may be parsed.

When aid_flag indicates that AI upscaling is not activated, syntax elements corresponding to AI sub data are not parsed.

The syntax element corresponding to the AI sub-data may include aid_average_qp, aid_qp_selection, aid_multiplication, aid_codec_information, aid_resolution_information, and aid_meta_flag, and may correspond to aid_average_qp, aid_qp_selection, aid_multiplication, aid_meta_resolutioninformation, aid.

aid_average_qp is a syntax element corresponding to aid_average_qp of FIG. 17 and may be information indicating an average value of a quantization parameter applied to an image. aid_qp_selection is a syntax element corresponding to aid_qp_selection of FIG. 17 and may be indicator information indicating a degree (or level) of quantization of at least a part of an image.

aid_multiplication is a syntax element corresponding to aid_multiplication of FIG. 17 and may be information indicating a ratio (or multiple) between the original image and the first image.

aid_codec_information is a syntax element corresponding to aid_codec_information of FIG. 17 and may be information indicating a predetermined codec used for first encoding of an image.

aid_resolution_information is a syntax element corresponding to aid_resolutaion_information of FIG. 17 and may be information indicating the resolution of the original image (or the first image).

aid_meta_flag is a syntax element corresponding to aid_meta_flag of FIG. 17 and may mean a flag indicating whether or not additional metadata related to AI upscaling is included.

When aid_meta_flag indicates that the AI upscale-related additional metadata is included (if(aid_flag)), syntax elements corresponding to the AI upscale-related additional metadata may be parsed.

The additional metadata obtained may include hdr_max_luminance, hdr_color_gamut, hdr_pq_type, rate_control_type, aid_target_bitrate, and the like. hdr_max_luminance is HDR (High Dynamic Range) maximum illuminance applied to high-resolution image, hdr_color_gamut is HDR color gamut applied to high-resolution image, hdr_pq_type is HDR PQ (Perceptual Quantizer) information applied to high-resolution image, rate_control_type is image obtained as a result of the first encoding A rate control type applied to data, aid_target_bitrate, may indicate a target bitrate. According to an embodiment, a specific syntax element among the syntax elements corresponding to additional metadata may be parsed.

The syntax structure table 2400 of FIG. 24 is only an example, and syntax elements corresponding to some of the AI data shown in FIG. 23 may be included in the syntax structure table 2400 . Also, a syntax element corresponding to AI data not shown in FIG. 23 may be included in the syntax structure table 2400 .

Earlier, when explaining with reference to FIG. 23, it was described as if the value of the syntax element was parsed in units of bytes, but this is to emphasize that the payload part including AI data is parsed in units of bytes in consideration of byte alignment. , the value of each syntax element may be parsed in units of byte, but is not limited thereto, and is parsed in various units (eg, by u(1) or ae(v) functions) based on a defined syntax structure table can be

However, the present invention is not limited thereto, and the syntax structure table applicable to the above-described AI data may be similar to the syntax structure table including the above-described AI data.

25 is a diagram for explaining operations of a bitstream-based AI encoding apparatus and an AI decoding apparatus according to an embodiment.

Referring to FIG. 25 , the AI encoding apparatus 1400 may receive an original image as an input and output a video bitstream to the AI decoding apparatus 1300 . For example, the AI encoding apparatus 1400 may determine whether the original image is AI downscaled, and may AI downscale the original image according to whether the AI downscale is performed to generate the first image. In this case, the AI downscaling may be performed by the AI downscaling unit 1412 . If AI downscaling is not performed, the original image may be the input image of the first encoder 1414 instead of the first image.

The first encoder 1414 may generate image encoding information by first encoding a first image based on a predetermined codec, and may express the image encoding information in the form of a bitstream. The image encoding information may be included in a main bitstream region defined by a predetermined codec, in which image encoding information in a bitstream may be included. For example, when a predetermined codec is one of MPEG H.264, HEVC, and VVC codecs, the video encoding information may be included in a NAL unit having the types of VPS, SPS, PPS, I slice, B slice, and P slice. have. On the other hand, when the predetermined codec is the AV1 codec, the image encoding information may be included in OBUs of which the types of OBUs are sequence header OBU, frame header OBU, frame OBU, and tile group OBU.

The first encoder 1414 may express AI data in the form of a bitstream based on a predetermined codec. The AI data may be included in a sub-bitstream region defined by a predetermined codec, in which additional data in the bitstream may be included.

AI data may include AI sub-data for selecting DNN information for upscaling used for AI upscaling corresponding to AI upscaling and an AI upscaling activation flag indicating whether upscaling corresponding to AI downscaling or not. . For example, AI data may be included in the NAL unit of the SEI message when a predetermined codec is one of MPEG H.264, HEVC, and VVC codecs. When the predetermined codec is AV1, AI data may be included in the metadata OBU.

The first encoder 1414 may generate a video bitstream including AI data expressed in the form of a bitstream and image encoding information.

The AI decoding apparatus 1300 receives the video bitstream output from the AI encoding apparatus 1400 . The first decoder 1332 may first decode image encoding information included in the video bitstream to obtain a second image. At this time, when the predetermined codec is an MPEG codec, based on the NAL unit type of the image encoding information included in the NAL unit header of the video bitstream, encoding information of SPS, PPS, I slice, P slice and B slice is obtained. and the obtained encoding information may be first decoded to obtain a second image.

The first decoder 1332 may acquire AI data included in the video bitstream. For example, when a predetermined codec is an MPEG codec, AI data may be obtained based on the NAL unit type of AI data included in the NAL unit header of the bitstream. That is, when the NAL unit type is an SEI message, the first decoder 1232 may acquire AI data from the corresponding NAL unit. In this case, the AI data may be included in an SEI message including user data registered by ITU-T T.35 or unregistered user data identified by UUID.

Meanwhile, when the predetermined codec is the AV1 codec, the first decoder 1332 may obtain AI data from the metadata OBU among the OBUs. In this case, the AI data may be included in the metadata OBU including user data registered by ITU-T T.35.

The AI decoding apparatus 1300 may determine whether to perform AI upscaling based on the AI upscaling activation flag included in the AI data.

The AI decoding apparatus 1300 may output the second image when the AI upscaling activation flag indicates that the AI upscaling is not activated.

The AI decoding apparatus 1300 may input the second image to the AI upscaling unit 1334 and the AI data to the AI setting unit 1333 when the AI upscaling activation flag indicates that the AI upscaling is activated. have.

The AI setting unit 1333 selects one upscale DNN information from among a plurality of upscaling DNN information based on the AI data, and the AI upscaling unit 1334 selects upscaling DNN information based on the selected upscaling DNN information. A third image may be output by performing AI upscaling on the second image based on the scale DNN.

26 is an exemplary diagram illustrating a mapping relationship between various pieces of image-related information and various pieces of DNN configuration information.

26 , it can be seen that the AI encoding/AI decoding process according to an embodiment of the present disclosure does not only consider a change in resolution. 26 , resolutions such as SD, HD, Full HD, 4K, bit rates such as 10Mbps, 15Mbps, 20Mbps, and 25Mbps, and codec information such as AV1, H.264, HEVC, and VVC, Sports,Movie DNN setting information can be selected individually or in consideration of video genre information such as , General, and QP level information. Here, the QP_level information may be the QP level indicated by aid_qp_selection of FIG. 17 (or the QP indicator of FIG. 21 ), but is not limited thereto, and the aid_qp_selection and aid_average_qp of FIG. 17 or a quantization parameter obtained when the image encoding information is first decoded It may be a range of the QP value determined by at least one of. An embodiment in which the DNN configuration information is selected in consideration of the QP level (or the range of the QP value) will be described later with reference to FIG. 22 .

For this consideration, training that considers each element in the AI training process should be connected in the encoding and decoding process.

Therefore, according to the training content, as shown in FIG. 26 , when a plurality of DNN setting information is provided based on image-related information including a codec type and image resolution, the AI decoding apparatus receives the information delivered in the AI decoding process. DNN setting information for AI upscaling of the second image 135 may be acquired based on information related to the first image 115 or the original image 105 . The information related to the first image 115 or the original image 105 may be obtained by first decoding image encoding information, but may also be obtained from AI data.

That is, the AI setting unit 1333 may use the DNN setting information according to the image related information by matching the image related information shown on the left side of the table shown in FIG. 26 and the DNN setting information on the right side of the table.

As illustrated in FIG. 26 , the resolution of the first image 115 is SD from information related to the first image 115 or the original image 105 , and image data generated as a result of the first encoding of the first image 115 . has a bit rate of 10Mbps, a ratio of the first image 115 to the original image 105 is 1.5x, the first image 115 is first encoded with the AV1 codec, and the genre of the first image 115 is If it is sports and it is confirmed that the QP level is 5, the AI setting unit 1333 may use A DNN setting information among a plurality of DNN setting information.

In addition, from the information related to the first image 115 or the original image 105_, the resolution of the first image 115 is SD, and the bit rate of image data generated as a result of the first encoding of the first image 115 is 10Mbps, the ratio of the first image 115 to the original image 105 is 1.5x, the first image 115 is first coded with the AV1 codec, the genre of the first image 115 is a movie, and QP When it is confirmed that the level is 5, the AI setting unit 1333 may use B DNN setting information among a plurality of DNN setting information.

Any one of A DNN setting information and B DNN setting information is selected according to whether the genre of the first image 115 is sports, a movie, or a general public. Under the same other conditions, the genre of the image may be different, and the first DNN and the second DNN may be jointly trained based on the genre of the image. DNN setting information can be obtained according to the genre of

Also, from the information related to the first image 115 or the original image 105, the resolution of the first image 115 is SD, and the bit rate of image data generated as a result of the first encoding of the first image 115 is 15 Mbps. , the ratio of the first image 115 to the original image 105 is 1.5x, the first image 115 is first encoded with the AV1 codec, the genre of the first image 115 is a movie, and the QP level If it is confirmed that this is 5, the AI setting unit 1333 may use the C DNN setting information among a plurality of DNN setting information.

Either one of B DNN setting information and C DNN setting information is selected according to the bit rate of the image data. Under the same other conditions, the bit rate of the image data may be different, and the first DNN and the second DNN may be jointly trained based on the bit rate of the image data. DNN setting information may be acquired according to the genre of the image 135 .

Also, from the information related to the first image 115 or the original image 105 , the resolution of the first image 115 is SD, and the bit rate of image data generated as a result of the first encoding of the first image 115 is 10 Mbps. , the ratio of the first image 115 to the original image 105 is 1.5x, the first image 115 is first encoded with the AV1 codec, the genre of the first image 115 is general, and the QP level If it is confirmed that this is 5, the AI setting unit 1333 may use the C DNN setting information among a plurality of DNN setting information.

Also, from the information related to the first image 115 or the original image 105 , the resolution of the first image 115 is SD, and the bit rate of image data generated as a result of the first encoding of the first image 115 is 10 Mbps. , the ratio of the first image 115 to the original image 105 is 2.0x, the first image 115 is first encoded with the AV1 codec, the genre of the first image 115 is sports, and the QP level If it is confirmed that this is 5, the AI setting unit 1333 may use D DNN setting information among a plurality of DNN setting information.

Either one of C DNN setting information and D DNN setting information is selected according to whether the ratio of the first image 115 and the original image 105 is 1.5x or 2.0x. The ratio of the first image 115 and the original image 105 may be different under the same other conditions, and the first DNN and the second DNN are based on the ratio of the first image 115 and the original image 105 . Thus, the AI setting unit 1333 may acquire DNN setting information according to the ratio of the first image 115 to the original image 105 .

Also, from the information related to the first image 115 or the original image 105 , the resolution of the first image 115 is SD, and the bit rate of image data generated as a result of the first encoding of the first image 115 is 10 Mbps. , the ratio of the first image 115 to the original image 105 is 2.0x, the first image 115 is first encoded with the AV1 codec, the genre of the first image 115 is sports, and the QP level If it is confirmed that this is 4, the AI setting unit 1333 may use E DNN setting information among a plurality of DNN setting information.

Any one of D DNN setting information and E DNN setting information is selected according to the QP level (or range) of the first image 115 . In other conditions, the QP level may be different, and the first DNN and the second DNN may be jointly trained based on the QP level. can be obtained

In addition, from the information related to the first image 115 or the original image 105 , the resolution of the first image 115 is HD, and the bit rate of image data generated as a result of the first encoding of the first image 115 is 15 Mbps. , the ratio of the first image 115 to the original image 105 is 1.5x, the first image 115 is first encoded with the H.264 codec, the genre of the first image 115 is sports, When it is confirmed that the QP level is 3, the AI setting unit 1333 may use F DNN setting information among a plurality of DNN setting information.

In addition, from the information related to the first image 115 or the original image 105 , the resolution of the first image 115 is HD, and the bit rate of image data generated as a result of the first encoding of the first image 115 is 15 Mbps. , the ratio of the first image 115 to the original image 105 is 1.5x, the first image 115 is first encoded with the AV1 codec, the genre of the first image 115 is general, and the QP level If it is confirmed that this is 3, the AI setting unit 1333 may use G DNN setting information among a plurality of DNN setting information.

Either one of F DNN setting information and G DNN setting information is selected according to the type of codec. Under the same other conditions, the type of codec may be different, and the first DNN and the second DNN may be jointly trained based on the type of the codec. Setting information can be obtained.

In addition, from the information related to the first image 115 or the original image 105, the resolution of the first image 115 is Full HD, and the bit rate of image data generated as a result of the first encoding of the first image 115 is 15Mbps, the ratio of the first image 115 to the original image 105 is 1.5x, the first image 115 is first coded with the AV1 codec, the genre of the first image 115 is general, and QP When it is confirmed that the level is 3, the AI setting unit 1333 may use H DNN setting information among a plurality of DNN setting information.

Any one of G DNN setting information and H DNN setting information is selected according to the resolution of the first image 115 . Under the same other conditions, the resolution of the first image 115 may be different, and the first DNN and the second DNN may be jointly trained based on the resolution of the first image 115 . Accordingly, the AI setting unit 1333 may acquire DNN setting information according to the resolution of the first image 115 .

In addition, from the information related to the first image 115 or the original image 105, the resolution of the first image 115 is Full HD, and the bit rate of image data generated as a result of the first encoding of the first image 115 is 20Mbps, the ratio of the first image 115 to the original image 105 is 1.5x, the first image 115 is first coded with the HEVC codec, the genre of the first image 115 is a movie, and QP When it is confirmed that the level is 2, the AI setting unit 1333 may use I DNN setting information among a plurality of DNN setting information.

Also, from the information related to the first image 115 or the original image 105, the resolution of the first image 115 is 4k, and the bit rate of image data generated as a result of the first encoding of the first image 115 is 25 Mbps. , the ratio of the first image 115 to the original image 105 is 1.5x, the first image 115 is first encoded with the VVC codec, the genre of the first image 115 is sports, and the QP level If it is confirmed that this is 2, the AI setting unit 1333 may use J DNN setting information among a plurality of DNN setting information.

Also, from the information related to the first image 115 or the original image 105, the resolution of the first image 115 is 4k, and the bit rate of image data generated as a result of the first encoding of the first image 115 is 25 Mbps. , the ratio of the first image 115 to the original image 105 is 1.5x, the first image 115 is first encoded with the VVC codec, the genre of the first image 115 is a movie, and the QP level When it is confirmed that this is 2, the AI setting unit 1333 may use K DNN setting information among a plurality of DNN setting information.

In another embodiment, the AI setting unit 1333 considers both the information (mode information, quantization parameter information, etc.) provided from the first decoder 1232 and the information related to the first image 115 included in the AI data. Thus, it is possible to obtain DNN setting information for AI upscaling of the second image 135 among a plurality of DNN setting information. For example, the AI setting unit 1333 receives quantization parameter information used in the first encoding process of the first image 115 from the first decoder 232 , and receives the quantization parameter information from the AI data of the first image 115 . The bit rate of the image data generated as a result of encoding may be checked, and DNN setting information corresponding to the quantization parameter and the bit rate may be obtained. Even at the same bit rate, there may be a difference in the quality of the reconstructed image depending on the complexity of the image, and the bit rate is a value representative of the entire first image 115 to be first encoded, and the bit rate of each frame in the first image 115 is The picture quality may be different. Accordingly, when mode information (eg, prediction mode information, motion information, etc.) and/or quantization parameters obtainable for each frame, frame group, and video from the first decoder 1232 are considered together, AI data only is used. In comparison, more suitable DNN setting information for the second image 135 may be obtained.

Also, depending on the implementation, the AI data may include an identifier of mutually agreed DNN configuration information. The identifier of the DNN setting information is an upscale target corresponding to the downscale target of the first DNN, and the DNN setting information trained in association between the first DNN and the second DNN so that the second image 135 can be AI upscaled. It is information for distinguishing pairs of . After acquiring the identifier of the DNN setting information included in the AI data, the AI setting unit 1333 may AI upscale the second image 135 by using the DNN setting information corresponding to the identifier of the DNN setting information.

Also, according to implementation, AI data may include DNN configuration information. After the AI setting unit 1333 obtains the DNN setting information included in the AI data, the AI upscaling unit 1334 may AI upscale the second image 135 by using the corresponding DNN setting information.

According to the embodiment, when information constituting the DNN setting information (eg, the number of convolution layers, the number of filter kernels for each convolutional layer, parameters of each filter kernel, etc.) is stored in the form of a lookup table, The AI setting unit 1333 obtains DNN setting information by combining some selected among the lookup table values based on the information included in the AI data, and the AI upscaling unit 1334 uses the obtained DNN setting information to obtain a second The image 135 may be upscaled by AI.

According to an embodiment, when the structure of the DNN corresponding to the upscale target is determined, the AI setting unit 1333 may obtain DNN setting information corresponding to the determined structure of the DNN, for example, parameters of the filter kernel.

In an embodiment, the AI setting unit 1333 sets a DNN for AI upscaling the second image 135 among a plurality of DNN setting information based on the resolution information of the third image 145 included in the AI data. You can also obtain information.

The AI setting unit 1333 obtains DNN setting information of the second DNN through AI data including information related to the first DNN, and the AI upscaling unit 1334 sets the second DNN setting information to the obtained DNN setting information. AI upscales the second image 135 through AI, which can reduce memory usage and computational amount compared to upscaling by directly analyzing the characteristics of the second image 135 .

In an embodiment, the AI setting unit 1333 considers at least one of the performance information of the display device, the setting value stored in advance in the AI decoding apparatus 1300, and the setting value input from the user, and selects the first one of the plurality of DNN setting information. 2 It is also possible to obtain DNN configuration information for AI upscaling of the image 135 .

For example, the AI setting unit 1333 may acquire DNN setting information corresponding to a setting value stored in advance in the AI decoding apparatus 1300 . For example, when a specific picture quality and/or resolution is previously stored in the AI decoding apparatus 1300, the AI setting unit 1333 generates a third image 145 having a specific picture quality and/or resolution stored in advance. It is possible to obtain DNN configuration information for

As another example, the AI setting unit 1333 may acquire DNN setting information corresponding to the setting value input by the user. For example, when a specific picture quality and/or resolution is input by the user, the AI setting unit 1333 may obtain DNN setting information for generating the third image 145 having the input specific picture quality and/or resolution. may be

As another example, the AI setting unit 1333 may include performance information (eg, resolution information of an image that can be reproduced by the display device, display DNN setting information for generating the third image 145 having a specific image quality and/or resolution may be acquired in consideration of the image quality information that the device can reproduce. For example, when the display device supports only FHD (Full HD) resolution of 1920x1080 as the maximum resolution and the resolution of the second image 135 is 1280x720, the AI setting unit 1333 sets the resolution of the second image 135 It is possible to obtain DNN configuration information that can increase .

In an embodiment, when the second image 135 consists of a plurality of frames, the AI AI setting unit 1333 may independently acquire DNN setting information for each predetermined number of frames, or common for all frames. It is also possible to obtain DNN configuration information.

27 is a diagram for explaining determining indicator (or level) information of a quantization parameter based on a representative QP value of at least a part of an image, according to an embodiment.

Referring to FIG. 27 , the AI encoding apparatus/decoding apparatus may determine a representative QP value of at least a part of an image. The representative QP value may be an important factor determining image quality. Here, at least a part of an image may refer to various image data units such as a frame subgroup (eg, a slice), a frame, a frame group, and a video sequence. The representative QP value is a QP value representing at least a portion of QP values of an image, and may generally be a QP average value, but is not limited thereto, and may be a value such as a median value.

As an example, the AI encoding device/decoding device can be distinguished into cases where the representative QP values are 0-17, 18-24, 25-31, 32-38, 39-45, and 46-51, and in the range of QP values The corresponding quantization group #1 to quantization group #6 may be determined. That is, the AI encoding apparatus/decoding apparatus may determine a plurality of groups by grouping the QP values.

For example, when the representative QP value is 30, the AI encoding apparatus 1400 may determine quantization group #3, determine the QP indicator value corresponding to quantization group #3 as 3, and AI-decode QP indicator information may be transmitted to the device 1300 . In this case, the QP indicator may be used as an element for determining one DNN configuration information among a plurality of pre-stored DNN configuration information. A method of determining one piece of DNN configuration information using a QP indicator will be described later with reference to FIG. 28 .

The above description has been made on the assumption that the number of quantization groups is six with reference to FIG. 27 , but the present invention is not limited thereto, and the number of quantization groups may vary. Also, the range of the QP value may be different for each quantization group. The QP indicator does not deliver the QP value as it is, but determines a plurality of quantization groups (or levels) by grouping (leveling) the QP values, and increases the amount of signaled information by explicitly signaling information on the quantization group. It is possible to obtain DNN setting information more suitable for at least a part of the second image 135 without increasing the number.

28 is an exemplary diagram illustrating a mapping relationship between a QP value and various pieces of DNN configuration information.

As illustrated in FIG. 28 , according to a QP indicator value indicating a quantization group of at least a part of an image included in AI data, the AI setting unit 1333 configures a plurality of DNN setting information (eg, A to G DNN settings). information), one of DNN configuration information may be determined.

Also, according to the range of the QP value to which the average QP value for the image included in the AI data belongs, the AI setting unit 1333 may determine one piece of DNN setting information among a plurality of DNN setting information for the image. In addition, the representative value of QP is determined based on the quantization parameter of at least a part of the image included in the image encoding information, and according to the range of the QP value to which the representative QP value belongs, the AI setting unit 1333 is applied to at least a part of the image. It is possible to determine one DNN configuration information among a plurality of DNN configuration information.

Previously, a value of a QP indicator indicating a quantization group (or QP level) of at least a part of an image included in AI data, a QP average value for an image included in AI data, or a quantization parameter of at least a part of an image included in image encoding information A method of determining one DNN configuration information among a plurality of DNN configuration information based on the QP representative value determined based on has been described, but the present invention is not limited thereto. Those skilled in the art can understand that one piece of DNN configuration information can be determined. In addition, those skilled in the art can understand that one piece of DNN configuration information among a plurality of DNN configuration information may be determined by combining at least a part of the first image or the original image related information and at least part of the QP related value information.

Meanwhile, the above-described embodiments of the present disclosure can be written as a program or instruction that can be executed in a computer, and the written program or instruction can be stored in a medium.

The medium may continuously store a computer-executable program or instructions, or may be a temporary storage for execution or download. In addition, the medium may be various recording means or storage means in the form of a single or several hardware combined, it is not limited to a medium directly connected to any computer system, and may exist distributed on a network. Examples of the medium include a hard disk, a magnetic medium such as a floppy disk and a magnetic tape, an optical recording medium such as CD-ROM and DVD, a magneto-optical medium such as a floppy disk, and those configured to store program instructions, including ROM, RAM, flash memory, and the like. In addition, examples of other media include recording media or storage media managed by an app store that distributes applications, sites that supply or distribute various other software, and servers.

Meanwhile, the above-described DNN-related model may be implemented as a software module. When implemented as a software module (eg, a program module including instructions), the DNN model may be stored in a computer-readable recording medium.

In addition, the DNN model may be integrated in the form of a hardware chip to become a part of the AI decoding apparatus 200 or the AI decoding apparatus 1300 described above. For example, a DNN model may be built in the form of a dedicated hardware chip for artificial intelligence, or as part of an existing general-purpose processor (eg, CPU or application processor) or graphics-only processor (eg, GPU). it might be

In addition, the DNN model may be provided in the form of downloadable software. The computer program product may include a product (eg, a downloadable application) in the form of a software program distributed electronically through a manufacturer or an electronic market. For electronic distribution, at least a portion of the software program may be stored in a storage medium or may be temporarily generated. In this case, the storage medium may be a server of a manufacturer or an electronic market, or a storage medium of a relay server.

In the above, the technical idea of the present disclosure has been described in detail with reference to preferred embodiments, but the technical idea of the present disclosure is not limited to the above embodiments, and those of ordinary skill in the art within the scope of the technical spirit of the present disclosure Various modifications and changes are possible by the person.

100: AI encoding device
200: AI decryption device
1300, 1350: AI decryption device
1400, 1450: AI encoding device

Claims (18)

at least one processor executing one or more instructions;
The processor is
Obtaining a video bitstream including a main bitstream including image data obtained as a result of the first encoding of the first image and a sub-bitstream including AI data,
obtaining a second image corresponding to the first image by first decoding image data included in the main bitstream;
obtaining AI upscaling activation flag information included in the AI data of the sub-bitstream, and determining whether to perform AI upscaling on the second image based on the AI upscaling activation flag information;
When it is determined to perform AI upscaling on the second image,
Upscaling DNN set according to selected upscaling DNN information from among a plurality of upscaling DNN information stored in advance based on at least one of at least a portion of the image data and at least one of the AI sub data included in the sub bitstream Through, to obtain an AI upscaled third image from the second image,
When the third image is obtained, output the third image,
When it is determined not to perform the AI upscaling on the second image, the second image is output,
The AI upscaling activation flag information is flag information indicating whether AI upscaling is performed or not performed on the second image. The method of claim 1,
AI sub data included in the sub bitstream includes image genre information indicating the genre of the image, average quantization parameter information indicating the average quantization value of the first image, and quantization indicating the quantization degree of at least a part of the first image. parameter indicator information, image resolution information indicating the resolution of at least one of the original image and the first image, image ratio information indicating the ratio of the original image and the first image, and the codec used for encoding the first image An AI decoding apparatus comprising at least one of codec information indicating, metadata flag information indicating whether additional metadata related to AI upscaling is included, and additional metadata information related to AI upscaling. 3. The method of claim 2,
The processor is
Among the plurality of upscaling DNN information based on at least one of the image genre information, the average quantization parameter information, the quantization parameter indicator information, the image resolution information, the image ratio information, and the AI upscaling related additional metadata information AI decoding apparatus, characterized in that the upscaling DNN is set as one upscaling DNN information. 4. The method of claim 3,
The quantization parameter indicator information may be index information indicating one quantization parameter group among a plurality of predetermined quantization parameter groups,
and the processor sets the upscaling DNN as upscaling DNN information corresponding to one quantization parameter group indicated by the index information among a plurality of upscaling DNN information. 3. The method of claim 2,
The quantization parameter indicator information is index information indicating a quantization parameter group including an (average) value of at least one quantization parameter among a subgroup in a frame, a frame, a frame group, and a video sequence. The method of claim 1,
and the processor sets the upscaling DNN as the upscaling DNN information selected from among a plurality of upscaling DNN information based on the quantization parameter information obtained from the image data of the main bitstream. . The method of claim 1,
The main bitstream is included in a predetermined main bitstream area in a video bitstream defined by a predetermined codec, and the sub-bitstream is included in a predetermined sub-bitstream area in a video bitstream defined by a predetermined codec Characterized AI decryption device. 8. The method of claim 7,
The sub-bitstream is an AI decoding apparatus, characterized in that the SEI (Supplemental Enhancement Information) message. 9. The method of claim 8,
The SEI message is an SEI message including user data registered by a predetermined standard or unregistered user data identified by a Universally Unique Identifier (UUID). 3. The method of claim 2,
The sub-bitstream is an AI decoding apparatus, characterized in that the metadata OBU (Open Bitstream Unit). A computer-recordable recording medium storing a video bitstream including AI-encoded data, the recording medium comprising:
The video bitstream is
a main bitstream including encoding information of a first image obtained by AI downscaling the original image; and
and a sub-bitstream including AI data for AI upscaling of a second image, wherein the second image is obtained by decoding encoding information of the first image included in the main bitstream. corresponding video,
The sub-bitstream is
AI upscaling activation flag information indicating whether AI upscaling is performed or not performed on the second image; and
When AI upscaling of the second image is performed, computer recordable record comprising AI sub data used to select upscaling DNN information from among a plurality of pre-stored upscaling DNN information media. at least one processor executing one or more instructions;
The processor is
Determines whether or not to downscale the original video by AI,
Based on the determination result, an AI downscaled first image is obtained from the original image using a downscale DNN,
Obtaining image data by first encoding the first image,
generating AI upscaling activation flag information indicating whether AI upscaling corresponding to AI downscaling of the original video;
When it is determined to perform the AI downscaling on the original image, generate AI sub data related to AI upscaling corresponding to the AI downscaling,
generating AI data including the AI upscale activation flag information and the AI sub data;
generating a video bitstream including a main bitstream including the image data and a sub bitstream including the AI data;
transmit the video bitstream;
At least one of at least a part of the image data and at least a part of the AI sub data related to the AI upscaling is information used to select DNN setting information for upscaling from among a plurality of pre-stored DNN setting information for upscaling,
The DNN for upscaling used for AI upscaling is set as the selected DNN for upscaling setting information,
The AI upscaling activation flag information is flag information indicating whether AI upscaling is performed or not. In the AI decoding method of an image,
obtaining a video bitstream including a main bitstream including image data obtained as a result of a first encoding of a first image and a sub bitstream including AI data;
obtaining a second image corresponding to the first image by first decoding image data included in the main bitstream;
Obtaining AI upscaling activation flag information included in AI data of the sub-bitstream,
determining whether to perform AI upscaling on the second image based on the AI upscaling activation flag information;
When it is determined to perform AI upscaling on the second image,
Upscaling DNN set according to selected upscaling DNN information from among a plurality of upscaling DNN information stored in advance based on at least one of at least a portion of the image data and at least one of the AI sub data included in the sub bitstream through, acquiring an AI upscaled third image from the second image;
outputting the third image when the third image is obtained; and
outputting the second image when it is determined not to perform the AI upscaling on the second image;
The AI upscaling activation flag information indicates whether AI upscaling is performed or not performed on the second image. In the AI encoding method of an image,
determining whether to downscale the original video by AI;
obtaining an AI downscaled first image from the original image using a downscaling DNN based on the determination result;
obtaining image data by first encoding the first image;
Generates AI upscaling activation flag information corresponding to whether or not to downscale the AI of the original video, and when it is determined to perform the AI downscaling on the original video, AI related to the AI upscaling corresponding to the AI downscaling generating sub data;
generating AI data including the AI upscale activation flag information and the AI sub data;
generating a video bitstream including a main bitstream including the image data and a sub bitstream including the AI data; and
transmitting the video bitstream;
At least one of at least a part of the image data and at least a part of the AI sub data related to the AI upscaling is information used to select DNN setting information for upscaling from among a plurality of pre-stored DNN setting information for upscaling,
The DNN for upscaling used for AI upscaling is set as the selected DNN for upscaling setting information,
The AI upscaling activation flag information is flag information indicating whether AI upscaling is performed or not. A computer-readable recording medium recording a program for executing the method of claim 13. at least one processor executing one or more instructions;
The processor is
Obtaining a video bitstream including image data and AI data obtained as a result of the first encoding of the first image,
obtaining a second image corresponding to the first image by first decoding the image data;
obtaining AI upscaling activation flag information included in the AI data, and determining whether to perform AI upscaling on the second image based on the AI upscaling activation flag information;
When it is determined to perform AI upscaling on the second image,
Based on at least one of at least a portion of the image data and at least a portion of the AI sub-data, through the upscaling DNN set according to the upscaling DNN information selected from among a plurality of pre-stored upscaling DNN information, the second image Acquire the AI upscaled third image from
When the third image is obtained, output the third image,
When it is determined not to perform the AI upscaling on the second image, the second image is output,
The AI upscaling activation flag information is flag information indicating whether AI upscaling is performed or not performed on the second image,
The AI decoding apparatus according to claim 1, wherein the AI data is included in a header of a predetermined data unit including an encoding parameter related to the image data. at least one processor executing one or more instructions;
The processor is
Obtaining a video file including AI-encoded data including image data obtained as a result of the first encoding of the first image and AI data,
obtaining a second image corresponding to the first image by first decoding the image data;
obtaining AI upscaling activation flag information included in the AI data, and determining whether to perform AI upscaling on the second image based on the AI upscaling activation flag information;
When it is determined to perform AI upscaling on the second image,
From the second image through the upscaling DNN set according to the selected upscaling DNN information from among a plurality of upscaling DNN information stored in advance based on at least one of at least one of at least a portion of the image data and at least one of the AI sub data Acquire the AI upscaled third image,
When the third image is obtained, output the third image,
When it is determined not to perform the AI upscaling on the second image, the second image is output,
The AI upscaling activation flag information is flag information indicating whether AI upscaling is performed or not performed on the second image,
The video file may include: a media data box including the image data; and a metadata box including metadata related to the image data, wherein the metadata box includes the AI data. at least one processor executing one or more instructions;
The processor is
Determines whether or not to downscale the original video by AI,
Based on the determination result, an AI downscaled first image is obtained from the original image using a downscale DNN,
Obtaining image data by first encoding the first image,
generating AI upscaling activation flag information corresponding to whether AI downscaling of the original video is performed,
When it is determined to perform the AI downscaling on the original image, generate AI sub data related to AI upscaling corresponding to the AI downscaling,
generating AI data including the AI upscale activation flag information and the AI sub data;
Transmitting a video file including AI-encoded data including the image data and the AI data,
At least one of at least a part of the image data and at least a part of the AI sub data related to the AI upscaling is information used to select DNN setting information for upscaling from among a plurality of pre-stored DNN setting information for upscaling,
The DNN for upscaling used for AI upscaling is set as the selected DNN for upscaling setting information,
The video file may include: a media data box including the image data; and a metadata box including metadata about the image data, wherein the metadata box includes the AI data,
The AI upscaling activation flag information is flag information indicating whether AI upscaling is performed or not.

Images