JPH0284895A - Picture encoding system - Google Patents

Abstract

PURPOSE:To attain dimensional vector quantization without preparing plural different dimensional vector quantizers and increasing the number of picture signals by executing the quantization as it is when the size of a block goes to be nXn picture elements. CONSTITUTION:When the picture signal is inputted to a block dividing part 2, the division of the block is executed. In sub sampling parts 3 and 7, 16X16 and 8X8 picture elements are sampled and the 4X4 blocks are generated. In comparators 6 and 10, a value to be determined in advance is compared with the value of change quantity in the present block to be obtained by maximum value detecting parts 4 and 8 and minimum value detecting parts 5 and 9 and a detected result for the change quantity of the block is sent to a quantizing data setting part 11. Then, the data block to be quantized is set. For the set data, at first, an average value 12 is obtained and the data are quantized 13 and subtracted from the picture signal of the present block. After that, dispersion 14 is obtained and quantization 15 is executed. Then, normalization 16 is executed and vector quantization 17 is executed.

Description

[Description of the field to which the invention pertains] The present invention relates to an image encoding method that encodes image signals with high efficiency, and in particular, a vector quantum method that quantizes a plurality of image signals at once. It is about transformation.

[Prior art]

Vector quantization has been actively researched and developed in recent years as it has characteristics that can theoretically achieve image distortion and rate limits. This means that one image signal is grouped into multiple pixels (this is called a block, and if the block is n x n pixels, n x
This is called n-dimensional vector quantization. ), the error between the block and a preset representative pattern is determined, and the block is replaced with the representative pattern with the smallest error. When applying vector quantization to an image signal, one problem is to select a vector dimension that matches the characteristics of the image signal. In order to achieve characteristics close to the signal distortion/rate limit with vector quantization, it is necessary to make the vector dimension sufficiently large.

However, increasing the vector dimension requires a huge amount of processing to determine the error with the representative pattern and to set one representative pattern itself, which is difficult to realize. In addition, since image signals have large differences in the local statistics of one screen, for example, the appropriate block size differs between the background part and the outline part of the image. For these problems, (a
) Prepare multiple vector quantizers of different dimensions, perform large block vector quantization where a large block vector quantizer is suitable, and perform small block vector quantization where a small block is suitable. Switch. (b) A reduced image is generated by applying a low-pass filter and subsampling to the image signal, and the reduced image is enlarged and the difference is obtained from the original signal. The reduced image and the difference image obtained in this way are vector quantized in the same dimension. In this way, the effects of large-dimensional vector quantization on the reduced image and small vector quantization on the differential image can be obtained.

[Problem to be solved by the invention]

However, the method (a) requires a huge amount of memory because vector quantizers of different dimensions are prepared, and also requires a huge amount of processing to obtain a representative pattern.

In addition, the method (b) has the problem that the same image signal is processed multiple times, resulting in an increase in the amount of code. 1 + (1/2
X1/2)=1, it was necessary to process 25 times as many image signals.

The present invention has been made to solve the above problems.

An object of the present invention is to provide a technique that can perform dimensional vector quantization without the need to prepare a plurality of vector quantizers of different dimensions, and without causing an increase in the number of image signals to be processed. There is a particular thing.

The above and other objects and novel features of the present invention will become apparent from the description of this specification and the accompanying drawings.

[Means to solve the problem]

In order to achieve the above object, the present invention converts the image signal into k×
a block dividing unit that divides into blocks each consisting of k pixels;
The block unit k, the amount of change in the image signal within the block (
This is an image encoding system equipped with a change amount detection unit that calculates the activity), a reduction processing unit that reduces one image signal to 1/m, and an n×n-dimensional vector quantizer, where k=
When the blocks are divided so as to satisfy n×2i, and the amount of change detected in each block by the change amount detection section is smaller than a preset value, the reduction processing section divides the vertical and horizontal n.
/k to form a block consisting of n×n pixels, which is quantized by the vector quantizer, and if the amount of change is larger than a preset value, the block is reduced to /k again.
A first stage process is performed to divide the blocks into blocks each consisting of 2×k/2 pixels, and then the change amount detection section detects the change amount within the block for each redivided block, and the first stage process is performed. Similarly, comparing the amount of change with a predetermined value,
After reduction, the second step of vector quantization or block division is performed, and the third to i-1 steps are performed in the same way, and the process proceeds to the first step, where the size of the block is determined. The most important feature is that it is equipped with a means for vector quantization when it becomes n×n pixels, and that i, k, m, and n are integers.

[Effect]

According to the above-mentioned means, the blocks are divided so that k=n×2i is satisfied, and if the amount of change detected for each block by the change amount detection section is smaller than a preset value, the reduction processing section The pixels are reduced to n/k in length and width to form a block consisting of n×n pixels and quantized by the vector quantizer, and if the amount of change is larger than a preset value, the block is resized to /2×k/ The first stage of processing is performed to divide into blocks each consisting of two pixels, and then the change amount within the block is detected by the change amount detection section for each re-divided block, and the change is detected in the same manner as the first stage processing. The amount is compared with a predetermined value, and after reduction, a second stage process of vector quantization or block division is performed,
Similarly, the third to i-1 stages are processed, and once the block size reaches n×n pixels, vector quantization is performed as is, so multiple vector quantizers of different dimensions are used. In addition, dimensional vector quantization can be performed without the need to prepare separate image signals, and without causing an increase in the number of image signals to be processed.

[Embodiments of the invention]

An embodiment of the present invention will be specifically described below with reference to the drawings.

In addition, in all the figures for explaining an example.

Components having the same function are given the same reference numerals, and repeated explanations thereof will be omitted.

FIG. 1 is a block diagram showing a schematic configuration of an image encoding system according to an embodiment of the present invention.

In FIG. 1, 1 is an input terminal, 2 is a block dividing section,
3 is a sub-sampling section, 4 is a maximum value (=ax) detection section, 5 is a minimum value (win) detection section, 6 is a comparison section (Th),
7 is a sub-sampling section, 8 is a maximum value (wax) detection section, 9 is a minimum value (win) detection section, and 10 is a comparison section (Th)
, 11 is a quantization data setting section, 12 is an average value calculation section, 1
3 is a quantizer (Q), 14 is a variance calculation circuit, 15 is a quantizer (Q), 16 is a normalization processing section, 17 is a vector quantization section (VQ), and 18 to 21 are output terminals.

In the image encoding system of this embodiment, as shown in FIG. 1, an image signal is inputted from an input terminal 1 to a block dividing section 2. In the block dividing section 2, each image signal is
Divide into 16.8 x 8.4 x 4 blocks. The sub-sampling unit of the sub-sampling unit 3 corresponding to the reduction processing unit sub-samples 16×16 pixels to generate a 4×4 block. A maximum value detection section 4 and a minimum value detection section 5 correspond to the change amount (activity) detection section. The maximum value detection section 4 detects the maximum value within the 16.times.16 block, and the minimum value detection section 5 detects the minimum value within the 16.times.16 block. The comparison unit 6 compares the predetermined value with the amount of change (
activity).

The value of this amount of change (activity) is small in flat parts such as the background, and large in contour parts. General k, since it is more efficient to process the flat part in large blocks,
As a result of the comparison in the comparison section 6, the amount of change (activit
For blocks determined to have a small value of y), the output data of the sub-sampling unit 3 may be quantized;
The subsampling unit 7 subsamples 8×8 pixels to generate a 4×4 block. In the maximum value detection section 8,
The maximum value within the 8×8 block is determined to be 8 by the minimum value detection unit 9.
The minimum value within each x8 block is detected. The comparator 10 compares the predetermined value with the amount of change (activi
ty). These subsampled 4×4 blocks, the 4×4 blocks of the original signal, and the detection results of the amount of change (activity) of these blocks are sent to the quantization data setting section 11, and the data to be quantized is Set up blocks. For the set data, first, the average value of the image signal within the block is determined by the average value calculating section 12, and then quantized by the quantizer 13. After subtracting the quantized average value from the image signal of the current block, the variance calculation unit 14 calculates the variance of the image signal within the block, and the quantizer 1
Quantize by 5. The normalization processing unit 16 normalizes the image signal using the quantized variance value, and then the vector quantization unit 1? Vector quantize with . The parts 12 to 16 are used to make the vector quantizer used in the vector quantizer 17 general-purpose. The output terminal 18 outputs the quantization result of the average value, the output terminal 19 outputs the quantization result of the variance, and the output terminal 20 outputs the vector index. A signal indicating the block division form is outputted to the output terminal 21 .

Examples of block division formats are shown in FIG. 2A (division example) and FIG. 2B (data order). In this case, the division form as shown in FIG. 2A can be expressed as "10001" by using the data order 1.2 and 3.4 in FIG. 2B. In the data indicating the division form, the first "1" indicates that the largest block has been divided, and the second to fourth "O" indicate that the first to third blocks of the re-divided blocks are the same. Indicates that no more divisions were made. The first "1" indicates that the fourth of the re-divided blocks has been re-divided, and if there are up to three stages of division as in this configuration example,
No further data is needed to indicate the division.

As can be seen from the above description, according to this embodiment, blocks of more dimensions can be processed using a vector quantizer of one type of dimension, and there is no increase in the amount of processed data. In the case of this embodiment, the data shown can be expressed with a small number of bits, 5 bits at maximum and 1 bit at minimum (when a 16×16 block is not redivided). for example.

Assuming that a 16×16 block is not redivided and the number of representative patterns for vector quantization is 256, -
A block can be encoded with 9 bits. This is 1/14 or less compared to 128 bits when a 4×4 dimensional vector is simply used. In addition, since quantization is performed in small blocks in areas such as contours, there is little deterioration in image quality. Therefore, by using the present invention, the vector quantization dimension of the image signal can be changed to match the local image signal statistics. One image can be quantized with good quality and coding efficiency.

In this embodiment, the block division is 16×16,
Although three stages of 8X8.4X4 were used, of course other stages may be used. Furthermore, although subsampling processing is used as the image signal reduction processing section, it goes without saying that other methods may be used.

For example, it is also possible to use a value averaged over (k/n)×(k/n) pixels, or to perform subsampling after applying a Laplacian filter. In addition, the maximum value (
s+ax)・minimum value (win) is used, but the dispersion of image signals within a block may also be used. Further, the present embodiment may be used as is for still image encoding or intraframe encoding, or the present invention may be applied to interframe encoding, for example, in combination with motion compensation processing, or to motion difference images. good.

The present invention has been specifically explained above based on examples, but
It goes without saying that the present invention is not limited to the embodiments described above, and can be modified in various ways without departing from the spirit thereof.

〔Effect of the invention〕

As explained above, according to the present invention, there is no need to prepare a plurality of vector quantizers of different dimensions, and
, it is possible to perform dimensional vector quantization without increasing the number of image signals to be processed.

[Brief explanation of the drawing]

FIG. 1 shows block categories, and FIG. 2A (example of division) and FIG. 2B (data order) show a schematic configuration of an image encoding system according to an embodiment of the present invention. FIG. 3 is a diagram showing a block division form of an image encoding method according to an embodiment of the present invention. In the figure, 1...input terminal, 2...block division section, 3
゜7... Subsampling section, 4, 8... Maximum value detection section, 5.9... Minimum value detection section, 6, 10... Comparison section (Th), 11... Quantized data Setting section, 12...
・Average value calculation unit, 13°15... quantizer (Q), 1
4... Variance calculation circuit, 16... Normalization processing unit, 17
. . . Vector quantization unit (VQ), 18 to 21 . . . Output terminal.

Claims (1)

[Claims] (1) A block dividing unit that divides the image signal into blocks each consisting of k×k pixels, a change amount detection unit that calculates the amount of change in the image signal within the block for each block, and a reduction of the image signal to 1/m. An image encoding method including a reduction processing unit that performs k=n×
2^i, and if the amount of change detected in each block by the change amount detection section is smaller than a preset value, the reduction processing section reduces it to n/k in length and width. A block consisting of ×n pixels is formed and quantized by the vector quantizer, and if the amount of change is larger than a preset value, the block is re-divided into k/2×k/
The first stage of processing is performed to divide into blocks each consisting of two pixels, and then the change amount within the block is detected by the change amount detection section for each re-divided block, and the change is detected in the same manner as the first stage processing. The amount is compared with a predetermined value, and after reduction, the second stage of processing is performed such as vector quantization or block division again, and the third to i-1 stages are similarly performed, Proceeds to the i-th stage of processing and the block size is n×
An image encoding method comprising means for vector quantizing the number of n pixels as they are, and wherein i, k, m, and n are integers.