JPS62188575A - Encoding and decoding device for vector quantizing - Google Patents

Abstract

PURPOSE:To grasp the summary of the entire of a picture at an early time of decoding and reproducing the picture and to encode so as to approach to the definite picture sequentially with the elapse of the time by connecting a vector quantizer in multistages and encoding and decoding stepwise. CONSTITUTION:An encoding part performs the multistage connection by using three stages of the vector quantizers, an input signal system 1 in which a picture signal is divided into blocks is quantized by the first stage vector quantizer 2. The index 3 based on the first stage quantizing result is stored in an index buffer 2, the buffer 12 has the output selected by an encoding output selector 13, the first stage index 3 is initiallly selected, then, the second stage index 7 is selected, and the third stage index 11 is finally outputted as encoding data 14. A decoding part receives the encoding data 14, and then, a change over circuit 15 selects the first stage vector quantizing and decoder 16, the first stage index is decoded to obtain the first stage output vector 17 and it is written in a frame memory 22.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a vector quantization encoding/decoding device used for transmitting/recording image signals, etc.

[Conventional technology]

FIG. 6 is a block diagram showing the configuration of a conventional vector quantization decoder disclosed in, for example, IEICE Technical Report IT85-61''High-Efficiency Image Coding by Vector Quantization''. 39 is an input vector register, 40 is a distortion calculation circuit, and 41 is an output vector address counter.

42 is an output vector code table, 43 is a minimum distortion detection circuit, 44 is an index strobe signal, 45 is an index register, and 46 is an index.

FIG. 7 is a block diagram showing the configuration of a conventional vector quantization decoder. In the figure, 46 is an index;
7 is an index register, 42 is an output vector code table, 48 is an output vector register, and 49 is an output vector.

First, we will briefly explain the principle of vector quantization. Now, the input signal series are put together into input vector quantity=(ut+u21..., U).

At this time, the K-dimensional Euclidean signal space RK (uERK
) of N representative points (i.e. output vectors) 2” (v
Let the set of ll, vl, , ..., vlK) be V = [ヱyo, ヱyo, . . . , ヱ□].

The vector quantizer selects from among the set of output vectors,
The closest distance to the input vector (least distortion)
The output vector Eyu is determined as follows, and it is searched.

if d (u, v,)(d (u, v*) f
or all j''L→ヱyu, where d (u, ヱyu) is the distance (distortion) between the input and output vectors. At this time, it is transmitted or recorded according to the index i of the input vector or output vector,
At the time of playback, it is replaced by the output vector UE.

The set of output vectors Eyu is obtained by clustering using the signal sequence serving as the training model (selection of representative points and quantization of each representative point of the training model are repeated until the total sum of distortion is minimized). be able to.

Next, the operation of the conventional vector quantization encoder/decoder will be explained with reference to FIGS. 6 and 7. First, the vector quantization encoder will be explained with reference to FIG. The input signal sequence is input to the input vector register 39 as an input vector 38.
be taken in. The output vector address counter 41 sequentially reads output vector 2 from the output vector code table 42. A distortion calculation circuit 40 calculates the intervector type of the input vector and the output vector X.

Next, the minimum distortion detection circuit 43 detects the minimum value of distortion between the output vectors sequentially read out from the distortion calculation circuit and the input vector quantity. When the minimum distortion detection circuit 43 detects the maximum distortion, an index strobe signal 44 is sent to the index register 45.

At this time, the indexer 46 of the output vector is taken into the index register 45, and the index 46 is output.

Next, the vector quantization decoder will be explained with reference to FIG. The input index 46 to be decoded is taken into the index register 47 and inputted into the output vector code table 42 to obtain a corresponding output vector, which is taken into the output vector register 48. The output vector 49 output from the output vector register 48 is decomposed into an output signal sequence to complete decoding.

When applying a vector quantizer to image encoding, the image is encoded using the correlation between neighboring pixels.
A method is used in which blocks are formed into blocks of n pixels (m and n are positive integers) and used as input vectors.

In the case of encoding, the blocks in one image are vector quantized one by one, and the encoding ends when the last block is quantized, and during that time, as many indexes as the number of blocks are output as encoded data. be done.

When decoding, the indexes are taken one by one, vector quantized and decoded to decode the blocks one by one, and the decoding ends when the index of the last block is decoded.

[Problem that the invention seeks to solve]

Conventional vector quantization encoding/decoding devices are configured as described above, so when applied to still image transmission or retrieval, when you want to select and obtain the necessary information, it is possible to Until the decryption and playback is complete, the entire contents cannot be grasped.
There is a problem in that if it takes time to transmit or reproduce a record, it is not possible to sort it immediately.

This invention was made to solve the above-mentioned problems, and it uses encoding in which the entire image can be seen at an early point in image reproduction, and as the decoding progresses, the image becomes more detailed. The purpose of this invention is to obtain a vector quantization encoding/decoding device that enables vector quantization.

[Means for solving problems]

The vector quantization encoding/decoding device according to the present invention includes:
Vector quantizers are connected in multiple stages, the latter stage vector quantizer quantizes the residual of the previous stage Vegtor quantizer, and the encoding/decoding process is performed sequentially from the previous stage.

[Effect]

The vector quantization encoding/decoding device according to the present invention performs encoding/decoding in vector quantizers connected in multiple stages.
As the decoding process is performed sequentially starting from the front-stage vector quantizer, the decoded and reproduced image gradually changes from a rough image with a lot of quantization distortion due to coarse quantization to an image with high definition. It can be determined whether the image information is necessary.

〔Example〕

An embodiment of the present invention will be described below with reference to the drawings. 1st
The figure is a block diagram showing the configuration of the encoding unit of the vector quantization encoding/decoding device for image encoding according to the present invention. In FIG. 2 is the first stage vector quantizer that encodes the input signal sequence 1, 3 is the first stage vector quantizer 2
4 is the locally decoded first stage output vector, 5 is the quantization error signal of the first stage vector quantizer 2, and this quantization error signal 5 is the input signal sequence 1.
a subtracter 5 which inputs the signal from , and the first stage output vector 4;
Obtained from 0.

6 is a second stage vector quantizer that quantizes the quantization error signal 5, 7 is a second stage vector: an index of the encoded output of the digitizer 6, 8 is a second stage output vector, and 9 is a second stage vector quantizer. This quantization error signal 9 is a quantization error signal of the second stage vector quantizer 6 and is obtained from a subtracter 51 which inputs the quantization error signal 5 and the second stage output vector 8. 10 is a third stage vector quantizer, 11 is a third stage vector quantizer 1
It is an index that is the encoded output of 0. The combination of the n vector quantizers and subtracters constitutes a hierarchical multi-stage vector quantizer.

12 is the index 3, 7 of each stage quantizer 2, 6.10
．． 11 is an index buffer for storing 11, 13 is a coding output selector, and 14 is coded data output from the coding unit as a coded output.6 FIG. 2 shows a vector quantization coding device for image coding according to the present invention. 2 is a block diagram showing the configuration of a decoding unit. In FIG. 2, 14 is encoded data to be decoded, and 15 is a vector quantization decoder 16, 18, 20 used for decoding, which will be described later. This is a switching circuit that switches according to the number of stages of the quantizer when quantization is performed in the encoding section, and this switching circuit 15 is provided on the input side and output side of the vector quantization decoder 16.18゜2o. There is. 16 is a first-stage vector quantization decoder, 1
7 is a first stage output vector, and 18 is a second stage vector quantization decoder.

19 is a second stage output vector, 20 is a third stage vector quantization decoder, 21 is a third stage output vector, 22 is a decoded image frame memory, 23 is a decoded image signal, 57 is the output of the switching circuit 15 and decoding 24 is an adaptive spatial filter whose smoothing characteristic changes according to the number of decoding stages; 25 is a decoded image signal output that has passed through the adaptive spatial filter 24; be.

Next, the operation will be explained. First, the encoding section shown in FIG. 1 will be explained. This coding section uses three stages of vector quantizers and performs multi-stage connection, and in this embodiment, the input signal sequence 1, which is a block of the image signal, is used as an input vector and is quantized by the first-stage vector quantizer 2. The vector in the first stage has 64 (8×8) dimensions. Index 3 based on the first stage quantization result is stored in the index buffer 12.

The first-stage vector quantizer 2 outputs an output vector 4 based on the quantization result, and obtains the first-stage quantization error signal 5 by taking the difference from the first-stage input signal sequence 1. The first stage quantization error signal 5 is quantized by the second stage vector quantizer 6,
A second stage index 7 and a second stage output vector 8 are output. Second stage index 7 is index buffer 1
It is stored in 2.

Further, a difference signal 9 between the second stage output vector 8 and the second stage input signal 5 series is obtained and used as an input signal to the third stage vector quantizer 10. The dimension of the vector in the second stage vector quantizer 6 is lower than that in the first stage to 16 (4×4). The third-stage vector quantizer 10 also outputs an output index 11 in the same way as the first-stage and second-stage vector quantizers 2.6. The dimension of the vector in the third stage vector quantizer 10 is assumed to be 4 (2×2).

The output of the index buffer 12 is selected by the encoded output selector 13, and outputs the index 3 of the first stage first, then the index 7 of the second stage, and finally the index 11 of the third stage as encoded data 14.

Next, the operation of the decoding section that decodes the encoded data obtained by the encoding section of FIG. 1 will be explained with reference to FIG. 2. When the decoding section receives the encoded data 14, the switching circuit 15 selects the first stage vector quantization decoder 16, the first stage index is decoded, the first stage output vector 17 is obtained, and the switching circuit 15 selects the first stage vector quantization decoder 16. The data is written into the frame memory 22 via the device 57.

The first stage output vector 17 is added to the frame memory output 23 by the adder 57 before the frame memory 22, but at the time of first stage decoding, the content of the frame memory at the corresponding position is zero, so the frame memory 22 The content of is the image formed by the output vector 17 of the first stage.

The decoded image signal 23 output from the frame memory has quantization distortion removed by an adaptive spatial filter 24 whose smoothing characteristics are switched depending on the number of decoding stages, and is output as a decoded image signal 25 to the display.

When the input of the index of the first stage vector quantizer 16 as encoded data 14 is completed and the second stage index arrives, the switching circuit 15 selects the second stage vector quantization decoder 18 and The index is decrypted. The output vector 19 of the second stage is combined with the decoded image signal 23 of the first stage already written in the frame memory 22 and the adder 5.
7 and the image with smaller quantization error is recorded in the frame memory 22.

When the second stage decoding is performed, the quantization distortion becomes smaller, so the smoothing characteristic of the adaptive spatial filter 24 is weakened compared to the first stage. In this way, the decoded image signal output 25 has better image quality than at the first stage.

When the second stage vector quantization decoding is finished and the third stage index arrives, the switching circuit 15 selects the third stage vector quantization decoder 18. Third stage output vector 2
1 is added to the encoding result 23 up to the second stage, which has already been written to the frame memory 22, and is written to the frame memory 22. The adaptive spatial filter 24 further weakens the smoothing characteristic than in the second stage, and the decoded image signal 25, which has been completely decoded, is output to the display.

The above is an example in which three types of vector quantizers, 64-dimensional, 16-dimensional, and 4-dimensional, are connected in multiple stages, but it goes without saying that other numbers of dimensions and stages can be used.

FIG. 5 is an explanatory diagram showing an example of the adaptive spatial filter 24 used in the decoding section. To express the filter characteristics of the adaptive spatial filter 24 in a formula, for the pixel arrangement shown in FIG. 5, if the filter output at the pixel position of X is X', then α: control parameter (0 × α≦1). An example is the characteristic of

In the above embodiment, when there is a lot of quantization distortion in the decoded output as in the case of first-stage vector quantization decoding, the control parameter α is set to a small value in order to strengthen the smoothing characteristic, and the third-stage vector quantization decoding As in decoding, when the decoding output has little quantization distortion and high resolution is desired, α is set to a value close to 1 in order to weaken the smoothing characteristic. In the second-stage vector quantization decoding, which is in the middle, it is better to set α to the middle. *By controlling the spatial filter 24 as described above, the image displayed during decoding can be quantized Distortion is reduced and the sequential playback process becomes more natural.

In the above embodiment, vector quantizers are provided in the same number as the number of quantization stages, but one quantizer can be used repeatedly to achieve the same effect as multi-stage connection.

FIG. 3 is a block diagram showing the configuration of the encoding section of the image signal vector quantization encoding device according to the present invention, in which an encoding loop is formed using a vector quantizer that can adaptively switch the number of dimensions. .

In FIG. 3, 26 is a variable block size blocking circuit that blocks the input signal sequence 1 according to the dimensions of the vector quantizer, and 27 is the residual obtained by subtracting the quantization results from the input signal vector up to the previous stage using the subtracter 52. 28 is a variable dimension vector quantizer, 29 is a stage reduction output vector code table, 30 is an index, 3
1 is an output vector based on quantization, 32 is an image vector of quantization results up to the number of stages of quantization, and this image vector is output from an adder 53 that adds the quantization results up to the previous stage and the output vector. . 33 is a variable block size frame memory, and 34 is an image vector of the quantization results up to the previous stage.

FIG. 4 is a block diagram showing the configuration of a decoding section to which the index signal 30 is input, and the same parts as in FIG. 2 are given the same reference numerals. In FIG. 4, 35 is a variable-dimensional vector quantization decoder, 54 is a stage-reduced output vector code table, 36 is an output vector, 55 is a variable block size frame memory, 37 is an image vector of the quantization results up to the previous stage, 56 is an adder that adds the output vector and the image vector.

Next, the operation will be explained. First, the encoding section shown in FIG. 3 will be explained. The input image signal from input signal series 1 enters the variable block size blocking circuit 26, where it is initially blocked in units of 8×8 pixels and converted into a 64-dimensional vector.

The vector 34 of the quantization results up to the previous stage is subtracted from the blocked vector by the subtracter 52, but since the content of the corresponding position in the frame memory 33 is initially zero,
The output of the blocking circuit 26 directly becomes the input vector 27 of the variable-dimensional vector quantizer 28.

The variable-dimensional vector quantizer 28 operates as a first-stage quantizer, and executes 64-dimensional vector quantization while quoting a code table for the first stage from the stage-reduced output vector code table 29. The variable dimension vector quantizer 28 outputs an index 30 of the first stage quantization and a first stage output vector 31, and the output vector is added to the output 34 of the frame memory 33 (zero in the first stage quantization) in the adder 53. After that, each 8×8 pixel is written into the block size variable frame memory 33 as the first stage quantization result 32.

After the encoding of one frame of the input image signal is completed as the first-stage quantization, the second-stage quantization process is subsequently started. The blocking circuit 26 blocks 4×4 pixels and outputs a 16-dimensional vector, and the subtracter 52 subtracts the first-stage quantization result vector 34 and uses the residual as the input vector 27 of the variable-dimensional vector quantizer 28 .

The variable-dimensional vector quantizer 28 quotes the second-stage code table from the stage-reduced output vector code table 29, executes the second-stage 16-dimensional vector quantization, and sets the second-stage index 30. An output vector 31 is output. The output vector 31 is added to the vector 34 of the previous stage quantization result in an adder 53, and the image of the quantization result up to the second stage is written into the variable block size frame memory 33.

Similarly, when the second stage quantization is completed over the entire screen,
A third stage quantization is performed in the same loop.

In the third-stage quantization, the block size of the blocking circuit 26 is 2×2 pixels to create a four-dimensional vector, and the variable-dimensional vector quantizer 28 uses the code table for the third stage to generate the third-stage quantization. The index 30 of the index is output. In this way, the indexes of the first stage quantization to the indexes of the third stage quantization are output in order so that they can be transmitted or recorded.

Next, the operation of the decoding section that decodes the encoded data obtained by the encoding section of FIG. 3 will be explained with reference to FIG. 4. The fetched index 30 is first decoded into a 64-dimensional vector by the variable-dimensional vector quantization decoder 35, which functions as a first-stage decoder, while quoting the first-stage code table from the stage-reduced output vector code table 54.

Output vector 3 of variable dimensional vector quantization decoder 35
6 is added to the frame memory output 37 of the frame memory 55 by the adder 56, but in the case of first-stage decoding, the content of the frame memory 55 is zero, so the output vector 36
is written as is in the frame memory 55 in the third
This is similar to the encoding section in the figure. The output of the frame memory 55 is smoothed by an adaptive spatial filter 24 with variable smoothing characteristics, and is output as a decoded image signal output 25.

Thereafter, second-stage and third-stage decoding are performed in the same manner as in the case of encoding. In this way, by decoding the index in the order of first stage → second stage → third stage, coarse quantization becomes finer quantization step by step, and image quality improves.

〔Effect of the invention〕

As described above, according to the present invention, by connecting vector quantizers in multiple stages and performing encoding and decoding in stages, an overview of the entire screen can be grasped at an early point in image decoding and reproduction. This has the effect of providing a vector quantization encoding/decoding device that enables encoding that gradually approaches finer images over time.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of the encoding section of a vector quantization encoding/decoding device according to an embodiment of the present invention, FIG. 2 is a block diagram showing the configuration of the decoding section, and FIG. 3 is a block diagram showing the configuration of the decoding section. A block diagram showing the configuration of the encoding section of a vector quantization device according to another embodiment of the present invention, FIG. 4 is a block diagram showing the configuration of the decoding section, and FIG. 5 is a block diagram showing the configuration of the decoding section. Explanatory diagrams of the spatial filter, FIGS. 6 and 7, are block diagrams showing the configurations of the encoding section and the decoding section, respectively, of a conventional vector quantization encoding/decoding device. 1 is an input signal sequence, 2, 6.10 are vector quantizers with different dimensions, 3, 7.11 are output indices of the vector quantizer, 4, 8 are output vectors, and 5.9 is a quantization error signal , 12 is an index buffer, and 13 is a coding output selector. 14 is encoded data, 15 is a switching circuit, 16, 18°2
0 is a vector quantization decoder, 17, 19.21 is an output vector, 22 is a frame memory, 23 is a decoded image signal, 24 is an adaptive spatial filter, 25 is a decoded image signal output,
26 is a block size variable blocking circuit, 27 is a residual signal vector, 28 is a variable dimension vector quantizer, 2
9.54 is the output vector code table for each stage number, 30 is the index, 31.36 is the output vector, 32 is the quantization result image vector, 33.55 is the block size variable frame memory, 34 is the frame memory output, 35 is the variable dimension A vector quantization decoder, 37 is a frame memory output, 50 to 52 are subtracters, and 57.degree. 53.56 is an adder. In addition, in the figures, the same reference numerals indicate the same or equivalent parts.

Claims (2)

[Claims] (1) The image input signal sequence to be encoded is divided into multiple blocks and used as input vectors, and a subtracter that subtracts the output vector from the input vector to obtain the quantization error is combined, and the quantization error signal is a hierarchical multi-stage vector quantizer in which n combinations of vector quantizers and subtracters each having a different dimension are connected in multiple stages so as to perform quantization with a vector quantizer in each stage; and the hierarchical multi-stage vectors connected in multiple stages. An index buffer stores the output index of the quantizer separately for each quantizer in each stage, and the index of each stage is changed so that the output of the index buffer is switched so that the index is read out sequentially starting from the index of the first stage vector quantizer. an encoding unit having an encoding output selector for selecting readout of the encoding unit; and a vector quantization decoder having the same number of dimensions and quantization characteristics as the n vector quantizers of the encoding unit, respectively. a switching circuit that switches the vector quantization decoder according to the number of stages of the quantizer when the encoded data to be decoded is quantized in the encoding section; an adder that obtains a decoded image signal of the decoding stage by adding the output vector decoded by the decoder; a frame memory that stores the output of the adder as a decoded image signal of the decoding stage; It is equipped with an adaptive spatial filter that removes high-frequency components from the memory output using a smoothing characteristic that changes depending on the number of decoding stages, and then outputs it to the display. A vector quantization encoding/decoding device comprising a decoding unit that sequentially reproduces the quantization results in image frame units. (2) A variable block size blocking circuit that blocks the image input signal sequence to be encoded according to the dimensions of the vector quantizer, and a first read/write variable block size frame that contains the image of the quantization results up to the previous stage quantization. a memory, a subtracter that subtracts the quantized image signal up to the quantization stage from the input block to obtain the quantization error up to the quantization stage, and inputs the quantization error up to the quantization stage. a variable-dimensional vector quantizer that performs vector quantization with a number of dimensions according to the number of quantization stages as a vector, and the first reading/writing process that adds the quantization output vector and the quantized image signal up to the quantization stage. An encoding unit having an adder that obtains a quantized image of the relevant quantization stage to be written to a block size variable frame memory, and an output that refers to the number of dimensions of vector quantization decoding according to the number of quantization stages of encoded data to be decoded. a variable-dimensional vector quantization decoder that changes a vector code table; a second read/write block size variable frame memory into which images of quantization results up to the decoding stage are stored;
The second read/write block size is determined by adding the decoded image up to the decoding stage read from the variable frame memory and the output vector decoded by the variable dimensional vector quantization decoder. It has an adder that obtains decoded images up to the relevant decoding stage to be written into a variable frame memory, and an adaptive spatial filter that changes smoothing characteristics according to the number of decoding stages, and has an A vector quantization encoding/decoding device comprising a decoding unit that sequentially reproduces quantization results in image frame units.