KR20060011281A - Apparatus for converting resolution of image applied to transcoder and method of the same - Google Patents

Abstract

Disclosed are a resolution converting apparatus and method applied to a transcoder. The cubic convolution interpolator is a second size continuous image data having a second resolution by a cubic convolution interpolation function modified to make the terminal gradient a fixed value of the decoded discrete image data of the first size having the first resolution. Convert to The sampling unit resamples the continuous image data by a predetermined size conversion coefficient. According to the present invention, it is possible to increase the quality of the reconstructed image, to reduce information loss generated during the size conversion process, and to perform reliable resolution conversion by an arbitrary size conversion coefficient.

Resolution Conversion, Cubic Convolution, Interpolation, Transcoder, MPEG

Description

Apparatus for converting resolution of image applied to transcoder and method of the same}

1 is a block diagram showing a detailed configuration of an embodiment of a transcoder to which a resolution converter according to the present invention is applied;

2 illustrates the available formats;

3 to 5 are views illustrating a process of converting a resolution from 1920 × 1080i to 720 × 480i, from 1920 × 1080i to 1280 × 720p, and from 1280 × 720p to 720 × 480i, respectively;

6 is a block diagram showing a detailed configuration of a preferred embodiment of a resolution converting apparatus according to the present invention;

7 is a diagram illustrating a process of converting an image of a progressive scan format into an image of a progressive scan format having different resolutions;

8 is a diagram illustrating a process of converting an image of an interlaced scan format into an image of an interlaced scan format having different resolutions;

9 is a diagram illustrating a process of converting a frame of one interlaced scan format into a frame of two progressive scan formats;

FIG. 10 is a diagram illustrating a process of resizing two progressive scan format frames to one interlaced scan format frame; FIG.

11 and 12 are diagrams showing experimental results of a conventional resolution converter method for an enlarged experiment and a resolution converter method according to the present invention;

13 and 14 illustrate simulation results of a "mobile and calendar" image and a "flower garden" image, respectively.

FIG. 15 is a diagram showing experimental results on the performance of a hybrid transcoder composed of the resolution converter method shown in FIGS. 4 and 5;

 FIG. 16 is a diagram illustrating performance evaluation results of a transcoder using a B-spline filter. FIG.

The present invention relates to a resolution converting apparatus and method applied to a transcoder, and more particularly, to a modified cubic convolution resolution converting apparatus for converting a variety of formats in a transcoder.

Transcoding is an important technique for various video communications over heterogeneous networks with different bandwidths. Channel characteristics are not taken into account when an image is pre-encoded and corrected in a video on demand server. This creates a lack of flexibility in the transmission of pre-encoded video streams over heterogeneous networks. An effective way to overcome this problem is to use a transcoder that provides a match between a precoded MPEG bit stream and a transport channel. For example, the transcoder receives, as input, a MPEG-2 bit stream that has been previously encoded at a high bit rate and generates another low bit rate bit stream that meets the new bandwidth limit. There are two types of transcoding: transcoding in the spatial domain and the frequency (DCT) domain. During processing in the spatial domain, the stored bit stream is first decompressed and downscaling is performed on the pixel domain. Next, the derived data is recompressed at the target bit rate. I. Shandbleh and M. Ghanbari have proposed a transcoding technique that can reduce the conversion time of temporal and spatial resolution. Their technique works directly on the DCT domain. C. Yim, M, A, and Isnardi presented an efficient method for resizing images in the DCT domain with a scaling ratio of 2: 1. N. Merhav has proposed a method for performing approximate linear operations on images in the compressed region using an incremental technique in a system that includes a 2: 1 ratio down sampling module. According to the above studies, resolution conversion is very important for maintaining high quality of transcoded images. The conventionally proposed transcoder as described above has a disadvantage in that the spatial resolution can be changed only by an integer ratio such as ½, ¼, and the like.

On the other hand, the resolution conversion module is called a scaler. Typical scaling transforms discrete data into a continuous model and resamples these functions on a new sampling grid. According to the sampling theory, the original signal can be perfectly reproduced from the sample by convolution by the sinc function. However, since the sinc kernel is reduced to infinity too slowly, it is difficult to physically implement the function. Therefore, approximation methods such as bilinear, bicubic, cubic splines, and higher-order B-spline operations of more than three orders have been proposed. To obtain better quality scaled images, SK Park et al. Adapted cubic convolution interpolation parameters to the frequency magnitudes of measured images to be processed, and G. Ramponi proposed a spatial variable technique that introduced the concept of warped distance. did.

An object of the present invention is to provide a resolution converting apparatus and method applied to a transcoder that can change the size of an image at an arbitrary ratio and support various sizes of images and formats.

In order to achieve the above technical problem, a resolution converting apparatus applied to a transcoder according to the present invention includes: a cubic modified to make a terminal gradient a fixed value of decoded discrete image data of a first size having a first resolution; A cubic convolution interpolation unit for converting the second sized continuous image data having the second resolution by a convolution interpolation function; And a sampling unit for resampling the continuous image data by a predetermined size conversion coefficient.

In order to achieve the above technical problem, a resolution converting method applied to a transcoder according to the present invention is modified to make a terminal gradient a fixed value of a first size of decoded discrete image data having a first resolution. Converting into second size continuous image data having a second resolution by a cubic convolution interpolation function; And resampling the continuous image data by a predetermined size conversion coefficient.

As a result, the quality of the reconstructed image can be improved, information loss generated during the size conversion process can be reduced, and the resolution can be reliably converted by an arbitrary size conversion coefficient.

Hereinafter, exemplary embodiments of a resolution converting apparatus and method applied to a transcoder according to the present invention will be described in detail with reference to the accompanying drawings.

The general image encoding algorithm defined in the MPEG-2 standard is known to support various sizes of images and formats. In order to handle images of various formats, the resolution converter according to the present invention considers the relationship between sampling positions of signals. In particular, the following changes are made to the resolution converter according to the invention to correct the phase of the interpolation kernel.

First, while the conventional transcoding system converts the size of the image into an integer ratio, the transcoding system to which the resolution converter according to the present invention is applied converts the size of the image at an arbitrary ratio from 1920 × 1080i to 704 × 480i. Second, in order to support various transcoding cases, a resampling position change method for various cases such as sequential scanning format, interlaced scanning format, sequential scanning format and interlaced scanning format conversion, etc. is proposed.

In the transmission of MPEG-2 bit streams over digital storage media and bandwidth-limited channels, bit rate reduction is an essential process to ensure storage efficiency and transmission quality. This bit rate reduction is performed by the transcoder. The transcoder reduces the bit rate by storing coding parameters such as the size of the quantization step and the spatial / temporal resolution.

1 is a block diagram illustrating a detailed configuration of an embodiment of a transcoder to which a resolution converter according to the present invention is applied.

Referring to FIG. 1, a transcoder 100 to which a resolution converter according to the present invention is applied includes a decoding unit 110, a resolution conversion unit 130, and an encoding unit 140.

The decoding unit 110 decodes the input video stream. The decoding unit 110 includes a variable length decoder (VLD) 112, an inverse quantization unit 114, an inverse DCT unit 116, a frame memory 118, a motion compensator 120, and an adder. Has (122). The input video stream is dequantized through the VLD 112 and decoded up to the DCT domain. Such data is compensated for by the motion compensator 120 after the inverse DCT is performed by the inverse DCT unit 116. The data for which compensation by the motion compensator 120 is completed is data on the pixel domain. Since the structure and operation of the decoding unit 110 are the same as those of the decoder of the existing MPEG video, detailed description thereof will be omitted.

The resolution converter 130 converts the size of the decoded image data. The converted image is input to the encoding unit 140 and re-encoded into a bit stream having a different bit rate.

The encoder 140 encodes and outputs a video whose resolution is converted. The encoding unit 140 includes a DCT / quantization unit 142, a variable length coder 144, an inverse DCT / inverse quantization unit 146, a frame memory 148, a motion estimation unit 150, a subtractor 152, and The bit rate controller 154 is provided.

The image input from the resolution converter 130 is input to the DCT / quantizer 142, and then quantized with DCT, and then context-based adaptive variable length coding in the variable length coder 144. )do. At this time, the input image is also input to the inverse DCT / dequantization unit 146 and (DCT + Q) ^-1 is performed. The image on which (DCT + Q) ⁻¹ is performed is optionally stored in the frame memory 148 after the block boundary is smoothed. The motion estimation unit 150 performs motion estimation with the reference image and the input image stored in the frame memory 158 to subtract the reference image from the input image according to whether the input image to be encoded is an inter frame or an intra frame. The reference image is optionally provided to the subtractor 152 based on the determination. The bit rate controller 154 adjusts the bit rate according to the channel condition so that quantization is adaptively performed.

On the other hand, motion vectors and mode data obtained at the decoder side are reused at the encoder side. Many transcoders are widely used to reuse input motion vectors. The transcoder 100 having the resolution converter according to the present invention converts the motion vector to the same size as the resolution conversion ratio of the screen in order to reuse the motion vector, and the transformed motion vector has a narrow search range. (E.g., ± 2 pixels).

Hereinafter, a description will be given of a system capable of processing various video streams including SDTV in MPEG-2 MP @ HL and MP @ ML. Such a system supports conversion to a format defined in the MPEG standard. Formats defined in the MPEG standard include 1920x1080i, 1280x720p, 704x480p, 704x480i, and 640x480p. 2 shows the available formats, i and p represent interlaced and progressive scan formats, respectively. The resolution of the image is converted by non-integer factors in the horizontal and vertical directions, and therefore, the resolution converter 130 should be designed to enable non-integer ratio conversion. On the other hand, the IPC converting the conventional interlaced scan format into a sequential scan format merely changes the resolution of the image in the vertical direction twice, and converts the integer ratio in the horizontal and vertical directions.

3 to 5 illustrate a process of converting a resolution from 1920 × 1080i to 720 × 480i, from 1920 × 1080i to 1280 × 720p, and from 1280 × 720p to 720 × 480i, respectively. When the original image is an interlaced scan format, a splitter is used to generate even and odd fields. When the reconstructed image is an interlaced scan format, a combiner is used to generate one frame from two field images. In FIG. 4, the resolution converter changes the resolution from 1920 × 1080i to 1280 × 720p. At this time, the image size is reduced in the horizontal direction but the resolution in the vertical direction is increased. Therefore, up / down resolution converters must be used separately for the transverse and longitudinal directions. In the conversion process illustrated in FIG. 4, one i frame is converted into two p frames, while in the conversion process illustrated in FIG. 5, two p frames are converted into one i frame.

에 대해 많은 보간 함수들이 다음과 같이 정의될 수 있다.Hereinafter, the design of the resolution converter will be described in detail. The resolution conversion process consists of applying the original discrete data to the continuous function and resampling this function at the new sampling position. Equally Spaced Sampled Data

Many interpolation functions can be defined as

는 대응하는 보간 함수이고,

는 보간 커널이다. 또한, x와 x_k는 각각 연속값과 이산값을 나타낸다. 이와 같은 특징을 갖는 보간함수에는 큐빅 스프라인(cubic spline)과 선형 보간 함수가 있다. 수학식 1에서 x_k는 보간 노드이고, c_k는 샘플링된 데이터

에 의존하는 파라미터이다. 보간 커널

는 컨볼루션과 유사한 연산에 의해 이산 데이터

을 연속 함수

로 변환한다. here,

Is the corresponding interpolation function,

Is the interpolation kernel. In addition, x and x _k represent a continuous value and a discrete value, respectively. Interpolation functions with these characteristics include cubic splines and linear interpolation functions. In Equation 1, x _k is an interpolation node, and c _k is sampled data.

This parameter depends on. Interpolation kernel

Discrete data by convolution-like operations

Continuous function

Convert to

According to classical Shannon's sampling theory, if f (x) is band-limited in the range of (−π, + π), the following equation is satisfied.

Here, the sinc function is defined as

와 sinc(x)로 대체된다. 수학적 계산에 있어서, 수학식 2로 표현되는 이상적인 보간 공식은 보간 커널 sinc(x)의 감 소가 느린 비율로 진행되므로 실질적이지 못하다. 따라서, 폴리노미얼 스프라인 보간에 의해 재건하는 것이 유력한 대안이다.That is, c _k and β (x) in Equation 1 are respectively

And sinc (x). In mathematical calculations, the ideal interpolation formula represented by Equation 2 is not practical because the reduction of the interpolation kernel sinc (x) proceeds at a slow rate. Thus, reconstruction by polynomial spline interpolation is a viable alternative.

The most practical approach is to estimate the value of each unknown pixel using a subset of the nearest neighboring pixels. In one example, in a method based on a spline function, known data samples influence a value determined by the inverse of the distance function.

를 이용가능한 데이터로 가정하고,

를 보간될 값이라 가정한다. 또한, 가장 가까운 이용가능한 픽셀은 좌표상에서 x_k와 x_{k +1}에 위치하며, 샘플링 격자의 간격이 이러한 데이터에 대해 하나라고 가정한다. 한편, x, x_k, 그리고, x_{k +1} 사이의 거리는 다음식으로 정의된다.

Is assumed to be the available data,

Let is the value to be interpolated. Also, the nearest available pixel is located at x _k and x _{k +1} on the coordinates, assuming that the spacing of the sampling grid is one for this data. On the other hand, the distance between x, x _k , and x _{k +1} is defined by the following equation.

Here, 0 ≦ s ≦ 1 and x _k ≦ x ≦ x _{k +1} .

를 얻기 위해 사용될 수 있는 많은 알고리즘 중에서 가장 간단한 것은 다음의 수학식과 같다.Interpolated pixels

The simplest of the many algorithms that can be used to obtain is

Another more complex but more effective method is bicubic. The problem with this approach is that the discontinuity of the slope at the end of the waveform results in magnitude ripple in the reconstructed function. This problem can be eliminated by creating a cubic convolution function that makes the end slope of the interpolation fixed. The cubic convolution interpolation function can be expressed by the following equation.

 Rifman and Bernstein set α to -1, so that, at x = 1, β (x) has the same slope -1, as with the sinc (x) function. Keys suggested setting α = −½, whereby the interpolation function is very close to the original unsampled image in terms of power series. In Equation 7, since β (x) is zero except for the (-2, 2) section, substituting Equation 4 and Equation 7 into Equation 1 is summarized as follows.

이다. 나아가, 이들은 임의의 계수(정수, 유리수, 또는 무리수)에 의한 보간에 사용될 수 있다. Keys가 설정한 바와 같이 α=-½이라 하면, 수학식 8은 다음과 같이 변경된다.Here, x _k ≤ x ≤ x _{k + 1} , s = xx _k , 0 ≤ s ≤ ₁ , and the interpolation function matches the data sampled at the interpolation node. In other words,

to be. Furthermore, they can be used for interpolation by any coefficient (integer, rational, or irrational). If? =-½ as set by Keys, Equation 8 is changed as follows.

In Equation 7, the coefficient α may be used as an adjustment parameter to obtain the most visible interpolation. Park et al. Adapted the parameter α to the frequency of the particular image to be processed.

6 is a block diagram showing a detailed configuration of a preferred embodiment of a resolution converter according to the present invention. The resolution converter according to the present invention uses a cubic convolution converter defined by Equation 9 because it performs better than a bilinear interpolator and is simpler than B-spline.

Referring to FIG. 6, the resolution converter 600 according to the present invention includes a cubic convolution interpolation unit 610 and a sampling unit 620.

를 생성하며, 샘플링부(620)는 새로운 격자 y_n상에서 연속함수

를 재샘플링한다. 본 발명에 따른 해상도 변환기(600)는 이산 데이터

를 다른 해상도를 갖는 크기변환된 데이터

으 로 변환한다. δ>1인 경우에 크기변환된 데이터

는 원래의 데이터

가 확대된 데이터에 해당한다. 이 때, δ는 다음식과 같이 정의된다.The cubic convolution interpolator 610 is a continuous function

The sampling unit 620 generates a continuous function on the new grid y _n .

Resample. Resolution converter 600 according to the present invention is discrete data

Resized data with different resolution

Convert to. scaled data when δ> 1

Is the original data

Corresponds to the enlarged data. At this time, δ is defined as follows.

Here, M and N are the size of the original image and the converted signal, respectively.

는 수학식 9에 의해 계산된다. 크기변환은 연속함수

를 크기변환 계수 δ로 재샘플링하는 과정이며, 다음식과 같이 표현된다.Interpolated Continuous Function

Is calculated by equation (9). Scaling is a continuous function

Is a process of resampling by the size conversion coefficient δ and is expressed as follows.

_k x _n ≤y ≤x and _{k +1, s = y n -x} k one time, Equation 11 is converted as follows.

It is simple to apply this technique to multidimensional data such as gray level images. The size conversion process by the resolution converter shown in FIG. 6 may be applied separately along each axis. That is, the size transformation may be performed on the vertical axis after the size transformation on the horizontal axis.

는 각각 원영상의 폭과 높이, 그리고 크기변환된 영상의 폭과 높이를 의미한다. 가로방향으로의 재심플링은

를 만족하는 크기변환 계수 에 의해 수행되며, 이를 수식으로 표현하면 다음과 같다.The resolution converter according to the present invention may be used to convert the size of the progressive scan format to the image of the progressive scan format having different resolution as shown in FIG. In Figure 7

Denotes the width and height of the original image and the width and height of the resized image, respectively. Re-sampling in landscape

It is performed by the size conversion factor that satisfies the equation.

와

은 각각 영상의 가로방향으로 주어진 이산 데이터와 크기변환된 데이터이다. 그리고, y_n은 재샘플링 위치이다.here,

Wow

Are discrete data and size-converted data given in the horizontal direction of the image, respectively. And y _n is the resampling position.

On the other hand, the size conversion in the vertical direction is performed by the following equation.

와

은 각각 영상의 세로방향으로 주어진 이산 데이터와 크기변환된 데이터이다. 그리고, y_n은 재샘플링 위치이다.here,

Wow

Are discrete data and size-converted data given in the vertical direction of the image, respectively. And y _n is the resampling position.

와 홀수필드

)로 분리된 후에 수행된다. 본 발명에 따른 해상도 변환기는 도 8에 도시된 바와 같이 비월주사포맷의 영상을 해상도가 다른 비월주사포맷의 영상으로 크기변환하는 데 이용될 수 있다. 도 8에서

는 각각 짝수 및 홀수필드에 대한 가로 및 세로방향의 데이터이다. 가로방향으로의 재심플링은 다음의 수학식에 의해 수행된다.In addition, when the resolution converter according to the present invention is applied to an image of an interlaced scan format, the calculation according to Equation 12 and Equation 15 is performed by the frame image having two fields (that is, an even field).

And odd fields

After separation). As illustrated in FIG. 8, the resolution converter according to the present invention may be used to convert an image of an interlaced scan format into an image of an interlaced scan format having different resolutions. In Figure 8

Are horizontal and vertical data for even and odd fields, respectively. Resampling in the lateral direction is performed by the following equation.

The resized fields are combined as shown in FIG. On the other hand, the resampling in the longitudinal direction is performed by the following equation.

The combined longitudinal data is then represented by the following equation.

에 대한 샘플링 위치

는 원프레임 데이터

에서

에 대응된다. 그리고,

에 대한 샘플링 위치

는 원프레임 데이터

에서

에 대응된다. 이는 원프레임 데이터

에서의 재샘플링 위치가

임을 의미하며, 이는 곧 재샘플링 위치가 동일한 간격이 아님을 의미한다. 따라서, 세로방향의 데이터가 크기변경될 때 재샘플링 위치는 불규칙성을 보상하기 위해 다음과 같이 수정되어야 한다.In Equation 21

Sampling location for

Is one-frame data

in

Corresponds to. And,

Sampling location for

Is one-frame data

in

Corresponds to. This is one frame data

The resampling position in

This means that the resampling positions are not equally spaced. Therefore, the resampling position should be modified as follows to compensate for irregularities when the longitudinal data is resized.

는 천이된 샘플링 격자이다.here,

Is the transitioned sampling grid.

가 된다. 샘플링 위치

는

, 즉, 프레임 영상의 세로방향으로 동일하게 이격된 점들인

를 구성 하므로, d는 다음의 수학식과 같이 정의된다.In the vertical axis of the one-frame image, the sampling positions of Equations 22 and 23 are

Becomes Sampling position

Is

That is, the same spaced points in the vertical direction of the frame image

Since d is formed, d is defined as in the following equation.

Therefore, the sampling grid in the longitudinal direction of the odd field should be modified as shown in equation (24). The combined transverse data is obtained by the following equation.

In addition, the resolution converter according to the present invention is used to resize one interlaced frame into two frames of progressive scan format as shown in FIG. At this time, the binding operation is not performed. The size conversion is performed by the following equations for even and odd fields.

이 세로축에서 볼 때

과

에 대응된다. 이 경우 결합 연산이 수행되지 않기 때문에 대응되는 위치

와

는

가 되어야 한다. 따라서, d는 다음의 수학식과 같이 정의된다.In each field data, the sampling positions of Equations 27 and 29 are one-frame data.

In this vertical axis

and

Corresponds to. In this case, the corresponding position because no join operation is performed.

Wow

Is

Should be Therefore, d is defined as follows.

In addition, a resolution converter according to the present invention is used to resize two progressive scan frames to one interlaced frame as shown in FIG. At this time, the splitter is not used. In this case, the size conversion is performed by the following equations.

는 원래의 순차주사포맷의 프레임의 세로축에서 볼 때

를 구성한다. 따라서, d는 다음의 수학식과 같이 정의된다.Sampling Positions in Equations 32 and 34

Is viewed from the vertical axis of the frame in the original progressive scan format.

Configure Therefore, d is defined as follows.

을 얻기 위해 수학식 11에 따른 종래의 해상도 변환기법에서는 한번의 곱셈이 수행되어져야 한다. 한편, 본 발명에 따른 해상도 변환기법에서 수학식 23의 y_n을 얻기 위해 한번의 곱셈과 한번의 덧셈이 요구된다. y_n의 산출을 제외하면 s의 계산, 필터 커널의 생성, 및 필터링의 수행을 포함하는 다른 단계에서는 종래의 해상도 변환기법과 본 발명에 따른 해상도 변환기법이 동일한 복잡성을 갖는다. 다음의 표에는 수학식 9, 수학식 11, 수학식 12, 수학식 22, 및 수학식 23에 의해 샘플

를 얻기 위해 요구되는 수학적 복잡도가 기재되어 있다.The calculation amount of the conventional resolution converter method and the resolution converter method according to the present invention includes the determination of the sampling position y _n represented by Equation 11 or 23, the calculation of the phase S of Equation 12, the construction of the filter kernel represented by Equation 9, And FIR filtering according to Equation (9). Sample value in the calculation of y _n

In the conventional resolution converter method according to Equation 11, one multiplication must be performed to obtain. Meanwhile, in the resolution converter method according to the present invention, one multiplication and one addition are required to obtain y _n of Equation 23. Except for the calculation of y _n , the conventional resolution converter method and the resolution converter method according to the present invention have the same complexity in other steps including calculation of s, generation of filter kernels, and performing filtering. The following table shows a sample by Equation 9, Equation 11, Equation 12, Equation 22, and Equation 23.

The mathematical complexity required to obtain is described.

step Conventional Techniques According to Equations 9, 11, and 12 Techniques of the invention according to equations 9, 12, 22, and 23 Calculation of yn in Equation 11 or 23 Calculation of S in Equation 12 Generation of Filter Kernel in Equation 9 FIR Filtering in Equation 9 P = 1 A = 1 P = 3, A = 2 P = 4, A = 3 P = 1, A = 1 A = 1 P = 3, A = 2 P = 4, A = 3 Sum P = 8, A = 6 P = 8, A = 7

라 하면, 영상의 크기변환에 대한 전체 복잡도는 다음의 식에 의해 계산된다.The size of the resized image

In this case, the overall complexity of the image size conversion is calculated by the following equation.

종래의 해상도 변환기법에 비해

에 더해 추가적인 덧셈연산이 필요함을 의미한다. 곱셈연산(P)은 덧셈연산(A)보다 훨씬 긴 CPU 시간을 소모하므로, 덧셈연산에 의한 복잡성의 증가는 중요성이 떨어진다.The resolution converter method according to the present invention

Compared to the conventional resolution converter method

In addition, this means that additional addition operations are required. Since the multiplication operation P consumes much longer CPU time than the addition operation A, the increase in complexity by the addition operation is less important.

Computer simulations using real images were performed to evaluate the performance of the resolution converter method according to the present invention. "mobile and calender" images, "flewer garden" images, and "susie" images were used in the experiment. In order to obtain an objective evaluation of the resolution converter method according to the present invention, the amount of information loss caused by continuous size conversion was checked. Various criteria such as peak-to-peak signal to noise ratio (PSNR), signal to noise ratio (SNR), mean squared error (MSE), etc. may be considered to calculate information loss. Among these, PSNR and MSE have been used in previous papers. Therefore, PSNR and MSE are used as criteria for calculating information loss in this experiment. PSNR is applied as a reference for the other computer simulations shown in FIGS. 12, 14, 15, and 16.

The experimental images are lowpass filtered and then reduced in half using the conventional resolution converter method and the resolution converter method according to the present invention, respectively. Next, interpolation by the coefficient δ = 2 is performed in the horizontal and vertical directions to restore the original image size. The following table lists the PSNR and MSE calculated for the original image using the "mobile and calender" image, the "flewer garden" image, and the "susie" image as the experimental image.

Experimental video Resolution converter method 704 × 480i⇒704 / 2 × 480 / 2i⇒704 × 480i 704 × 480i⇒490 × 350i⇒704 × 480i Mobile and Calendar Conventional PSNR = 19.77㏈ MSE = 684.19 PSNR = 22.16㏈ MSE = 394.69 The present invention PSNR = 20.65㏈ MSE = 559.40 PSNR = 23.61㏈ MSE = 282.89 Flower garden Conventional PSNR = 21.02㏈ MSE = 513.55 PSNR = 23.82㏈ MSE = 269.73 The present invention PSNR = 21.69㏈ MSE = 440.04 PSNR = 24.67㏈ MSE = 221.72 Susie Conventional PSNR = 34.19㏈ MSE = 24.76 PSNR = 37.24㏈ MSE = 12.28 The present invention PSNR = 35.30㏈ MSE = 19.19 PSNR = 38.35㏈ MSE = 9.50

Referring to Table 2, the PSNR of the image interpolated by the resolution converter method according to the present invention is larger than the PSNR of the image enlarged by the conventional resolution converter method. In another experiment, the experimental image is reduced to a 490 × 350i image, and then the size conversion is performed to restore the original image size. In many cases, the resolution converter method according to the present invention showed better performance than the conventional resolution converter method. The PSNR trend in experiments for "mobile and calendar" images is repeated in simulations for "flower garden" and "susie" images. This is due to the fact that the conventional resolution converter method is not concerned with the relational information, while the interpolation kernel of the resolution converter method according to the present invention has been modified according to the relationship between the format of the sized picture and the original picture. These experimental results indicate that the resolution conversion algorithm according to the present invention has a significant improvement in terms of minimizing information loss when compared with the conventional interpolation technique. Accordingly, it can be appreciated that the resolution converter method according to the present invention is quite effective in the size conversion of the image resolution.

In other simulations for showing qualitative evaluation of the resolution converter method according to the present invention, experimental results of the conventional resolution converter method for the enlargement experiment and the resolution converter method according to the present invention are shown in FIGS. 11 and 12. The size conversion ratio is 5.7, and the resulting image is a sequential scan format. 11 and 12 illustrate noise areas of fences and horses in "mobile and calendar" images. Four consecutive pictures are combined into one image to compare the resulting images. Several horizontal lines are added in the figure to emphasize the differences. Whereas the horizontal lines of the enlarged image by the conventional resolution converter method do not coincide in the vertical direction, the resolution converter method according to the present invention generates an image such that the size-converted image is vertically regular in data.

In order to check the usefulness of the resolution converter method according to the invention in transcoding, the performance of the transcoder shown in FIG. 1 was evaluated. The video codec used for the simulation is MPEG-2 MP @ ML. The original image sequence is coded at a bit rate R ₀ (= 15M or 10M bps) and the size of the image is 704 × 480i. After the bit stream is decoded, the video sequence is resized to a 480x336i video. The scaled image is then transcoded at a bit rate R _T (= 10M or 5M bps). This simulation is one of the processes of converting a frame of an interlaced scan format as shown in FIG. 3 into a frame of another non-circular scan format. In order to evaluate image quality by SPNR, a size conversion process is used for reconstruction of an original image having a size of 704 × 480i. 13 and 14 show simulation results of the "mobile and calendar" image and the "flower garden" image, respectively. 13 and 14, the resolution converter method according to the present invention produces an average PSNR of 0.7 dB and 0.8 dB on average for each image as compared with the conventional resolution converter method.

In order to evaluate the performance of the transcoder having the resolution conversion process shown in FIGS. 4 and 5, the process of transcoding the image of the interlaced scan format shown in FIG. It combines the process of transcoding an image in a format into an interlaced image. The original image sequence ( "mobile and calendar" picture) is coded with a bit rate R _O = 15M bps, the size of the image is 704 × 480i. The encoded bit stream is transcoded into a 480 × 360p bit stream having a bit rate R ₁ = 10M bps, and then the transcoded 480 × 360p picture is transcoded again into a 704 × 480i picture having a bit rate R ₂ = 5M bps. FIG. 15 shows experimental results of the performance of the hybrid transcoder composed of the resolution converter method shown in FIGS. 4 and 5. Referring to Fig. 15, the resolution converter method according to the present invention produces an average PSNR of 0.7 mW and 1.2 mW on average compared with the conventional resolution converter method.

The performance evaluation of the transcoder using the B-spline filter to show that the phase shifter method according to the present invention can be applied to the B-spline filter is shown in FIG. The B-spline filter is used as a spatial resolution converter when other simulation conditions are maintained as shown in FIG. According to the simulation results, it can be seen that the resolution converter method according to the present invention produces a bit stream having a better quality than the conventional resolution converter method. This means that the resolution converter method according to the present invention can be usefully applied to a transcoder for changing the spatial resolution as well as the bit rate of an image sequence.

The invention can also be embodied as computer readable code on a computer readable recording medium. The computer-readable recording medium includes all kinds of recording devices in which data that can be read by a computer system is stored. Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage, and the like, and may also be implemented in the form of a carrier wave (for example, transmission over the Internet). Include. The computer readable recording medium can also be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion.

Although the preferred embodiments of the present invention have been shown and described above, the present invention is not limited to the specific preferred embodiments described above, and the present invention belongs to the present invention without departing from the gist of the present invention as claimed in the claims. Various modifications can be made by those skilled in the art, and such changes are within the scope of the claims.

According to the resolution converting apparatus and method according to the present invention, the quality of the reconstructed image can be improved by changing the size of various images defined in the MPEG standard by the modified cubic convolution resolution converter, and generated during the size conversion process. Information loss can be reduced. In addition, a reliable resolution conversion is possible by an arbitrary size conversion factor.

Claims (6)

Cubic convolving converting the decoded discrete image data of the first size having the first resolution into continuous image data of the second size having the second resolution by a cubic convolution interpolation function modified to make the terminal slope fixed. A friction interpolation unit; And And a sampling unit for resampling the continuous image data by a predetermined size conversion coefficient. The method of claim 1, The cubic convolution interpolation function is a resolution converter applied to a transcoder, characterized by the following equation:

, here,

Is the interpolated pixel value,

Is the available pixel value, and

to be. The method of claim 1, The sampling unit is applied to a transcoder, characterized in that for performing the resampling by applying the continuous image data to the following equation:

, here,

Is an interpolated pixel value, y _n is a resampling position, x is an available pixel position, and δ is a size conversion coefficient expressed as a ratio of the size of the scaled image to the size of the original image. Converting the decoded discrete image data of the first size having the first resolution into continuous image data of the second size having the second resolution by a cubic convolution interpolation function modified to make the end slope a fixed value; And Resampling the continuous image data by a predetermined size conversion coefficient; resolution conversion method applied to a transcoder comprising a. The method of claim 4, wherein The cubic convolution interpolation function is a resolution conversion method applied to a transcoder, characterized by the following equation:

here,

Is the interpolated pixel value,

Is the available pixel value, and

to be. The method of claim 4, wherein A method for converting a resolution applied to a transcoder, characterized in that resampling is performed by applying the continuous image data to the following equation in the sampling step:

, here,

Is an interpolated pixel value, y _n is a resampling position, x is an available pixel position, and δ is a size conversion coefficient expressed as a ratio of the size of the scaled image to the size of the original image.

Images