JPH03256453A - Compressed data amount control system - Google Patents

Abstract

PURPOSE:To control the amount of compressed data to a desired data amount by compressing the data while changing the value of a variable for each block, and following-up the accumulated data amount to a data amount function. CONSTITUTION:A calculated data amount function Gs is compared with an accumulated value Vs of the real compressed data amount for each block and in the case of Gs<Vs, a value subtracting a control value DELTAk from a variable (k) is defined as the variable (k) of the next block. In the case of Vs<Gs, a value adding the control value >=k to the variable (k) is defined as the variable (k) of the next block and in the case of Vs=Gs, the variable (k) is maintained for the next block as it is. In such a way, the value of the variable (k) is set for each block and the compressed data amount accumulated value Vs follows up the functions Gs while setting a coefficient satisfying a prescribed relation in the one-dimensional sequence of the transform coefficient to zero. Thus, the compressed data amount is controlled to the desired data amount.

Description

[Detailed Description of the Invention] [Industrial Application Field] This invention compresses still image data using a variable length code and then transmits or records the data so that the amount of data after compression becomes the required amount of data. This invention relates to a method for controlling the amount of compressed data.

[Conventional technology]

In order to standardize the natural image encoding method, “Ba5eli
'ne System' or 'Extended System'
Various international standardization methods such as em' have been proposed.

Figure 10 shows ``Ba5elin'' of the international standardization system.
This system divides one input image into multiple blocks of 8×8 pixels, and performs two-dimensional discrete cosine transformation (DCT) on each block. C
o51neTransform+g) (processing PI)
, quantization is performed by dividing the obtained DCT coefficients by each threshold of a quantization matrix consisting of 8 × 8 thresholds (
Processing P2). FIGS. 11 and 12 are examples of quantization matrices for luminance signals and color difference signals.

For the direct current (DC) component of the quantized DCT coefficients, the DCC bone difference quantized in the previous block is taken, and the number of bits of the difference is Huffman encoded.

For one minute of alternating current (AC), zigzag scan is performed within the block to convert it into a one-dimensional number sequence, and two-dimensional Huffman encoding is performed using the number of bits of effective coefficients and the number of consecutive zeros (invalid coefficients). P3 and P4).

FIG. 13 shows an example of a zigzag scan table.

Note that during quantization in process P2, each threshold value of the quantization matrix is multiplied by a certain coefficient (scale factor), and then the DCT coefficient is divided. The image quality and compression rate of the compressed image are adjusted by this scale factor.

The data compressed in this way is decompressed by the reverse process of processes P1 to P4. That is, Huffman decoding in process P5, DCC bone segment and ACC bone segment decoding in process P6, inverse quantization in process P7, and inverse DCT (10CT) in process P8.

[Problems to be Solved by the Invention] By the way, in the above-mentioned system, data compression is performed using a Huffman code, which is a variable length code, so the total amount of data after compression is the same as that at the end of the compression process (processes P1 to P4). You won't know until you do. Therefore, if it is necessary to perform encoding within a preset data amount range, some kind of data amount control is required.

This invention provides compressed data that controls the amount of data after compression to be a desired amount of data, and also compresses the image with the peripheral area of the image taken into consideration to reduce the amount of data. The purpose is to provide a quantity control method.

[Means for Solving the Problems] This invention allows one digital image to be divided into one block n×
Divide into multiple blocks consisting of n1i elements, perform discrete cosine transformation for each block, and obtain n×n
Among the transform coefficients, a predetermined one of the transform coefficients is set to zero according to the position of the block in the screen, and this n×
Quantization is performed by dividing n transform coefficients by each threshold of a quantization matrix consisting of n×n thresholds, and the quantized transform coefficients are divided in a fixed order from the DC component to the high frequency component. Set a specific value k for the one-dimensional sequence a, (p=o, I, ·, n2-1) obtained by converting it into a one-dimensional sequence, and After setting the coefficient a that satisfies the condition "q≦n2-1" to zero, data compression is performed to encode the number of consecutive zeros in the sequence ap, and the ACC bone sum of squares of the discrete cosine transform coefficients is A cumulative distribution is obtained for each block, and from this cumulative distribution, a function G that is proportionally converted to the desired compressed data amount is determined, and this function G and the cumulative value V of the actual compressed data amount obtained by data compression are combined into blocks. The values of the variables are adjusted for each block so that the cumulative value V follows the function G.

[For production]

In this invention, the adjustment of variables for each block is performed as follows. First, block 5 (s-I2, . . .
T), the sum of the squares of the discrete cosine transform coefficients Fu9OAC components is calculated. Next, calculate the cumulative distribution B for all blocks of this ACC bone sum of squares, and further calculate the cumulative distribution B
The data amount function G5, which is proportionally converted so that the maximum value BT of , corresponds to the desired data amount after compression, is determined by the following equation.

The data amount function Gs obtained in this way and the cumulative value V5 of the actual compressed data amount are compared for each block, and if C,<V, the value obtained by subtracting the adjustment value Δk from the variable k is set as the variable for the next block. If V<C, the value obtained by adding the adjustment value Δk to the variable k is set as the variable for the next block, and if V,=Gs, it is left unchanged.

In this way, the value of the variable is set for each block S, and "k≦q≦n" in the one-dimensional sequence ap of the transformation coefficient F ay is set.
-1'' is set to zero, and the compressed data amount cumulative value V is made to follow the function Gs.

In addition, the conversion coefficient of high frequency components is set to zero for each block according to its position on the screen, and data compression processing is performed with the periphery of the image essentially blanked out, so that the data compression processing is performed with the periphery of the image in a state where it is virtually blank. A large amount of data can be allocated to the central area, allowing efficient image compression that takes visual characteristics into consideration.

〔Example〕

FIG. 1 is a schematic diagram showing an embodiment of the processing procedure of the compressed data amount control method according to the present invention, and the same parts as in FIG. 10 are given the same reference numerals and will be explained.

First, input image data is divided horizontally and vertically into a plurality of χ×y blocks each consisting of n×n pixels, for example, 8×8 pixels, and a two-dimensional discrete cosine transform (DCT) is applied to each block. (processing Pi).

DCT is a type of orthogonal transformation in the frequency domain.
...+n-1), however, it is defined as C,-1/(2(w=0) = 1 (W≠0>).The transformation coefficient F ay transforms l blocks of human image data into spatial Indicates the component decomposed into frequency, and the coefficient F
0° is a value (DCC component) proportional to the average value of n×n pixels of the input image data f jj, and as u and v become larger, it represents a component with a higher spatial frequency (AC component).

Next, among the conversion coefficients F uv, a process is performed in which the coefficient F uv having the relationship “l≦(1+v”) is set to zero (process P10).The variable l is a value corresponding to the position of the block on the screen. (Process P11). Figure 2 is a diagram showing changes in the value of the variable l in each block on one screen. In this example, the block located in the center of the screen is set to 'l = 14', and the other blocks are is "l-4". Figure 3 shows the transformation coefficient F.v (Figure (a)) of a certain block.
F,,=O (however, 4≦u+v) (Figure (b))
It is a figure which shows the example which performed. As is clear from this figure, the processing in the processing PIO is a process in which components parallel to the diagonal of the transform coefficient indicated by the specified variable l and higher frequency components are forcibly set to zero, which is a kind of low-pass filter. Corresponds to processing. Therefore, when the variable 1 is small, the cutoff frequency of the low-pass filter is low, and as the variable 1 becomes large, the cutoff frequency becomes high.

FIG. 4 is a diagram showing another example of the variable l in each block on one screen. In this way, the value of the variable l is reduced for the peripheral part of the image that is not visually important to reduce high frequency components, thereby reducing the amount of data after compression.

Returning to FIG. 1, the DCT coefficients subjected to the low-pass filter processing (processing P10) are temporarily stored in the memory for each screen (processing P12). The stored coefficients are the sum of squares of AC components expressed by the following equation for each block.

(s=1.2.-, xXy) is taken (processing P13)
.

Next, the cumulative distribution B of the sum of squares for all blocks on one screen is expressed by the following equation for each block.

(process P14), B, = Σ b.

Furthermore, the maximum value of the cumulative distribution B, that is, the ACC status 2
The total sum B, r (T= x
Calculate the data amount function G, which is proportionally converted so that X y ) corresponds to the predetermined data amount after compression (process P15
).

FIG. 5 shows an example of the sum of squares 8, the cumulative distribution B, and the data amount function G.

In addition, the DCT coefficients for one screen stored in the memory are read out block by block and quantized by dividing each threshold value of a quantization matrix consisting of n×n threshold values by a scale factor (processing P2). Multiplication processing using a scale factor is performed by bit-shifting each threshold value of the quantization matrix.

Next, for the DC component, a difference is taken from the component quantized in the previous block (processing P30), and the number of bits of the difference is Huffman encoded (processing P4).

For ACIfi, zigzag scan is performed in the order shown in Fig. 13 to obtain the one-dimensional sequence al, (ρ-1゜2,
..., 63) (process P31). Figure 6 shows the matrix of DCT coefficients after quantization, and Figure 7(a)
shows the one-dimensional sequence ap after zigzag scanning.

Next, a process is performed on this sequence ap in which the value of the coefficient a9 that satisfies the relationship ``k≦q≦63.'' is set to zero (process P3
2), FIG. 7(b) shows the sequence a' after the process P32 is applied to the sequence a9 as "k-8". As is clear from FIGS. 6 and 7, the process P32 is a process for zeroing out the high frequency components after the second in the sequence a, and corresponds to a type of low-pass filter process. Therefore, by changing the value of k, the number of consecutive zeros can be changed, and the amount of data after compression, that is, the compression ratio can be adjusted.

Next, the number of consecutive zeros is run-length encoded (process P33), and two-dimensional Huffman encoding is performed using the run-length encoded data on the number of consecutive zeros and the number of bits of the effective coefficient (process P4).

Huffman encoding does not use the quantized coefficient values themselves for the DC component and ACC component, but performs Huffman encoding on the number of bits necessary to express the values.

In addition to the Huffman code, the value of the number of bits is added as additional information. For example, if the quantized coefficient is 2(1
0 base), if expressed in binary, it would be “000...
・010'', but the number of bits necessary to express this, 2, is Huffman encoded as a value representing this value, and 2
Bit data “°lO” is added as an additional bit.

On the other hand, if the quantized coefficient is negative, data obtained by subtracting 1 from the additional bit is added. For example, if the quantized coefficient is −
2 (decimal number), when expressed in binary number (2's complement representation), it becomes "111...110", and the lower two bits are additional bits, but "1" is subtracted from "10". 01'' is added as an additional bit. By doing this, when the quantized coefficient is positive, the additional bit starts with 1, and when it is negative, it starts with 0, making it easy to determine whether it is positive or negative.

Next, the amount of compressed image data for one block of compressed image data is counted and processed P16), and the cumulative amount of data V is calculated.
, and the data amount function G obtained in process P15 are compared (process P17), and the variables in the low-pass filter process are adjusted (process P18).

That is, if C,<V, the variable k is set to "k-Δk" to increase the number of consecutive zero data and reduce the amount of data after compression, and conversely, V,<G.

If so, the variable k is set to "k+Δk" to reduce the number of consecutive zero data and increase the amount of data after compression, and if V,=G, the data compression process for the next block is performed without changing.

This series of processing is repeated for all blocks, and the 8th
As shown in the figure, the variable k is controlled for each block so that the cumulative data amount after compression, ■, follows the data amount function G, and becomes the desired data amount.

To decompress compressed data, conventional data decompression processing may be performed. That is, Huffman decoding is first performed (processing P5), and for the DC component, differential decoding and ACt
After performing run-length decoding for 2 minutes, the data is rearranged in the order of zigzag scan to obtain transform coefficients for l blocks (processing P6), and a scale factor is assigned to each threshold value of the quantization matrix to these transform coefficients. The multiplied value is multiplied to perform inverse quantization (processing P7), and further, inverse discrete cosine transform (IDCT) is performed (processing P8), and the decompression process ends. In this case, the variables used during data compression are no longer needed during data decompression, so decompression processing can be performed using the conventional method.

Next, the operation of the data compression process will be explained in detail with reference to the flowchart shown in FIG.

First, initial settings of the system are performed (step S1). That is, the cumulative distribution B5.2 of the ACC bone sum of squares and 5.2 of the sum of squares bs is reset for all blocks, and the multi-values of the data amount function G are reset, and the block number is Set the value of S to the initial value "1".

Next, the image data of the first block is input (step 32), and DCT processing is performed (step S3). Next, a low-pass filter process is performed in which the transform coefficient F uv having the relationship "l≦u+v" is set to zero using a variable l corresponding to the position of the block on the screen (step S4).

Through this processing, the amount of data is reduced in advance for images in the periphery of the screen that are not visually important.

The transformation coefficient F uV obtained by the DCT processing and the low-pass filter processing is temporarily stored in the memory (step S
5). Then, the sum of squares b3 of the ACC bones is calculated for each block (step S6), and its cumulative distribution B is determined (step S7). In this case, since the first block (S=1) is being processed, B1 =b.

Next, determine whether all blocks have been processed (
Step S8). In this case, since the processing of all blocks has not yet been completed, "1" is added to the block number S (step S9), the image data of the next block is manually processed (step S2), and the steps 33 to S9 described above are performed. Repeat the process.

When all blocks have been processed, a data amount function G is calculated using the block number S as a variable (step 5IO), and an initial value ko is set as a variable for low-pass filter processing (step 5ll). The DCT coefficients of the first block are read out from the DCT coefficients for the screen and quantized (step 512).

If the coefficient after quantization is a DC component (step 513)
, the difference is taken from the DC component quantized in the previous block, and the difference value is encoded (step 5I4).

If the coefficient after quantization is an AC component (step 513)
, perform a zigzag scan to obtain a one-dimensional sequence a.

(step 315), performs low-pass filter processing to zero the high frequency components after the second in this sequence a (step 516), and performs run-length encoding to compress the number of consecutive zeros (step 517). .

Next, for the DC component, the encoded difference value is Huffman encoded, and for the ACC component, the number of consecutive zeros data that has been run-length encoded and the number of bits of the effective coefficient are
Perform Huffman encoding of the dimensions (step 318).

Then, it is determined whether all blocks have been processed (step 319), and if all blocks have been processed, the compression process is ended. In this case, since only the first block has been processed, the cumulative amount of data after compression V,
The process moves on to the process of determining (step 520). and,
The obtained cumulative data tvs and the data amount function G obtained in step SIO.

(step 521), if cs < Vs, set the variable k to "k-Δk" and increase the number of consecutive zero data (step 522); if VS < Gs, set the variable k to "k+Δk" and increase the number of consecutive zero data. Reduce the number of zero data (step 523), V, = G.

If so, data compression processing for the next block (steps 312 to 518) is performed using the same values.

In this way, data compression is performed while changing the value of the variable for each block, and as shown in FIG. 5, the cumulative data amount V3 follows the data amount function G, and the data amount after compression is is the desired amount of data.

Note that instead of using a constant value as the amount of change Δk in a variable, for example, if we use Δk - α (vs cs) (α: constant), then the error between the data amount function G and the actual data amount Vs This results in a feed puncture, which further improves tracking performance.

In addition, in this embodiment, the low-pass filter processing (processing P32) during the data compression processing is replaced by the quantization processing (processing P2).
Although it is performed after the quantization process, the same effect can be obtained even if the process is performed before the quantization process.

(Effect of the invention) According to the invention, a matrix of two-dimensional DCT coefficients is
Convert the DC component to a one-dimensional sequence in a fixed order from the DC component to the high frequency component, and calculate the DC of the high frequency component after the arbitrary number.
A method is used to adjust the amount of data by setting the T coefficient to zero, and the amount of data after compression is determined in advance for each block of the image to determine a data amount function, and the determined data amount function and the actual The data is compressed by comparing the cumulative amount of data for each block obtained by compressing the data, and performing data compression while changing the value of the variable for each block so that the actual compressed data amount follows the data amount function. It becomes possible to control the subsequent data amount to a desired data amount. In this case, by setting a predetermined conversion coefficient of each block to zero according to the position of the block on the screen, the periphery of the image is essentially left blank, and then the data compression process described above is performed. , it is possible to allocate a large amount of data to the central part of the screen, which is visually important, and to perform efficient image compression that takes visual characteristics into consideration.

Also, when determining the data amount function, the DC of each block
Since it is based on the sum of the squares of the AC components of the T coefficient, it is possible to allocate a large amount of data to areas where the image signal changes rapidly, and a small amount of data to areas where the changes are gradual, thereby reducing the image content. Natural compression can be performed according to the situation.

Furthermore, since variables used during data compression are not required during data expansion, data expansion can maintain compatibility with conventional methods.

In addition, if the variable k is changed not by a fixed range of change but by a value corresponding to the error between the actual compressed data amount and the data amount function, the actual compressed data amount can follow the data amount function. Performance is further improved.

[Brief explanation of drawings]

FIG. 1 is a diagram showing an embodiment of the processing procedure of the compressed data amount control method according to the present invention, FIGS. 2 and 4 are diagrams showing the value of the variable l in one image, and FIGS. (b) is a table showing the DCT coefficients before and after low-pass filter processing using the variable l;
Multiply and sum5. Cumulative distribution B8. A diagram showing an example of the data amount function Gs, FIG. 6 is a matrix table showing the DCT coefficients after quantization, FIG. 7 is a table converting the DCT coefficients into a one-dimensional sequence, and FIG. 8 is the data amount function and the actual Figure 9 is a flowchart to explain the data compression operation in Figure 1, Figure 10 is a diagram showing the processing procedure of conventional compression/expansion processing, Figure 11 is luminance. Table showing the quantization matrix of the signal, 1st
FIG. 2 is a table showing a quantization matrix of color difference signals, and FIG. 13 is a table showing a zigzag scan table.

Claims (2)

[Claims] (1) Divide one digital image into multiple blocks each consisting of n×n pixels, perform discrete cosine transform for each block, and select the relevant one among the n×n transform coefficients obtained by the transform. A predetermined transform coefficient among the above transform coefficients is set to zero according to the position of the block in the screen, and these n×n transform coefficients are divided by each threshold value of a quantization matrix consisting of n×n threshold values. The quantized transform coefficients are converted into a one-dimensional sequence a_p (p=0 , 1, ..., n^2-1), and set a specific value k for "k≦q in the above sequence a_p.
After setting the coefficient a_q that satisfies the condition "≦n^2-1" to zero, data compression is performed to encode the number of consecutive zeros in the sequence a_p, and the AC component of the discrete cosine transform coefficient is squared. The cumulative distribution of the sum is determined for each block, and from this cumulative distribution, a function G that is proportionally converted to the desired compressed data amount is determined, and this function G is combined with the cumulative value V of the actual compressed data amount obtained by compressing the data above. The compressed data amount control method is characterized in that the value of the variable k is adjusted for each block so that the cumulative value V follows the function G. (2) Let the adjustment value Δk of the variable k for each block be Δ
2. The compressed data amount control method according to claim 1, wherein k=α(V-G) (α: constant).