JPH05252535A - Picture data compressing device and method thereof - Google Patents

Abstract

PURPOSE:To improve a picture quality by appropriately distributing the remainder of generated code amounts. CONSTITUTION:A difference between the total of calculated predicted values and the target generated code amounts is searched, the remainder is divided into the number of entire blocks, and equally distributed to each block. Moreover, the remainder is divided into the number of blocks of a Y, and equally distributed only to the blocks of the Y, the remainder is divided into the number four times as many as that of the blocks in a V direction, and distributed to the most left side blocks of Y, R-Y, and B-Y in the ratio of 2:1:1, the remainder is equally distributed to the most left side blocks of the Y and the remainder is added to the left-and-up edge side block of the Y. Thus, the predicted generated code amounts of each block can be obtained.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a data compression apparatus used for image data compression processing, and more particularly to an image data compression apparatus having a generated code amount control function for controlling a generated code amount.

[0002]

2. Description of the Related Art As an international standard for image compression, JPEG (Jo
int Photographic Expert Group) and MPEG (Moving
Picture Expert Group). JPEG is intended to compress still images, and a color still image encoding method has already been determined and will be approved as an international standard. Regarding JPEG, a chip has been commercialized, and boards using this chip are beginning to appear on the market. The JPEG algorithm is roughly divided into two compression methods. The first method is DCT (Discrete Cosine).
Transform: A method based on Discrete Cosine Transform. The second method is DPCM (Differntial) in a two-dimensional space.
PCM) is a Spatial (spatial function) method. DCT
The method is lossy encoding that does not completely reproduce the original image because it includes quantization, but it is the basic method of this algorithm that can obtain sufficient decoded image quality even with a small number of bits. .. On the other hand, the Spatial method
It is a reversible encoding that has a small compression rate but completely reproduces the original image, and is a method added as a standard method to realize this characteristic.

The DCT method further includes a baseline process (Baseline System) which is an essential function and an extended DCT process (Extended System) which is an optional function.
Classified into one. In addition to these methods, there is a hierarchical process that combines the above methods to achieve progressive buildup. The baseline process is a minimum function that all encoders / decoders that implement the DCT method must have, and is an algorithm based on the ADCT method (Adaptive Descrete Cosine Transform Coding). Is. In the image compression in the baseline process, image data is processed in blocks of 8 × 8 pixels. The treatment process is as follows. (1) Two-dimensional DCT conversion process (2) DCT coefficient quantization process (3) Entropy coding process In the two-dimensional DCT conversion process, spatial data is converted into frequency data, and 64 DCT coefficients are output. At this time,
The color components are (Y, CB, CR). Of these coefficients, the coefficient at the upper left of the matrix is called the DC component and is the average value of the block data. The remaining 63 coefficients are called AC components.

In the quantization processing of the DC component, the DCT coefficient is linearly quantized by using a quantization matrix having a quantization step size of different magnitude set for each coefficient by a quantizer. However, in order to control the code amount or the decoded image quality, a value obtained by multiplying a quantization matrix by a coefficient (scaling factor) specified from the outside is used as an actual matrix value for quantization. In this way, the 64 DCT coefficients are quantized into integer values with reference to the table. This quantization process results in lossy compression. The contents of the reference table used are not specified in JPEG. The quantization table is created in consideration of human visual characteristics. Since humans are not sensitive to visual information of high frequency components, this high frequency component is roughly quantized.

In the entropy coding process, first, the difference between the DC component and the quantized DC component in the left adjacent block is calculated and coded. This method is called DPCM. The AC component is converted into a one-dimensional array by zigzag scanning. The Huffman coding method is used in the entropy coding of the baseline process. The Huffman coding process determines whether each coefficient is zero, and the coefficient of consecutive zeros is counted as a run length in its length. When a non-zero coefficient comes, two-dimensional Huffman coding is performed by combining the quantization result and the run length of the zero coefficient up to that point. The Huffman coding of the DC / AC coefficients is based on the given Huffman code table, but the quantization matrix and Huffman code table have no default values in order to be optimal in the situation in which they are used, if necessary. It is transferred from the encoder to the decoder for use.

FIG. 10 is a block diagram of a conventional image data compression apparatus 10 based on the JPEG algorithm.
In FIG. 10, reference numeral 11 denotes a frame memory, which holds original image data. The image data stored in the frame memory 11 is output to the DCT calculation unit 12,
The data is read, a DCT operation is performed in block units, and the result is output to the quantizer 13. The output data of the DCT operation is
The quantizer 13 is read by the quantizer 13 (FIG. 11).
Is a quantization table of the quantization table RAM 14 (see FIG.
2 and FIG. 13), the quantization operation is performed according to the value.
The Huffman coding unit 14 performs Huffman coding on the data output from the quantizer 13 using the DC code amount memory 15 and the AC code amount memory 16, and stores the total generated code amount in the total generated code amount storage memory. Store in 17. CPU1
Reference numeral 8 controls the entire apparatus, and also controls the generation of an optimum quantization table, which will be described later, and the prediction of the generated code amount of each block. The Huffman encoder 14 performs Huffman encoding on the data output from the quantizer 13. The operation of the Huffman encoder 14 has the following two types of modes, mode1 and mode2. (Mode 1) Huffman coding is performed on the input data, and the number of bits generated by the DC component of each block is written in the DC code amount memory 15. Also,
The number of bits generated by the AC component is written in the AC code amount memory 16. (Mode2) Huffman coding is performed on the input data, and at that time, the predicted value of the generated code amount due to both the DC component and the AC component of each block written in advance in the AC code amount memory 16 and the actual generation are generated. If the generated code amount is larger than the predicted amount, the EOB (End of
Block) is generated and the coding of the block is completed. The surplus number of bits is added to the predicted value of the next block.

FIG. 11 is a block diagram of the quantizing device 13. 11, the quantization device 13 includes a quantization table RAM 21 for storing a Y (luminance information) quantization table (FIG. 12) and a quantization table RAM 21 for storing a C (color information) quantization table (FIG. 13). Table RAM22 and D
Each frequency component calculated by CT is multiplied by a value in a quantization table (that is, multiplied by an inverse number and divided) to perform quantization for changing the weighting. In addition, the quantization table RAMs 21 and 22
Corresponds to each of the Y, RY, and BY components.

FIG. 12 is a Y quantization table, and FIG. 13 is C.
It is a figure which shows the quantization table for use, and the reciprocal of a quantization table is actually written in said quantization table RAM21,22. The quantizer 13 outputs this multiplication result. With the above configuration, the image data compression apparatus 10 performs the following operations to predict the optimum quantization table and the generated code amount in each block. i) First, an appropriate quantization table is created and written in the quantization table RAMs 21 and 22. ii) Next, DCT → quantization → Huffman coding (mo
Perform de1). iii) The total generated code amount held in the AC code amount memory 16 is referred to, a new appropriate quantization table is created based on the value, and the created quantization table is written in the quantization table RAMs 21 and 22. To be crowded. iv) Then, DCT → quantization → Huffman coding (mo
Perform de 1). v) A quantization table that generates a target code amount is created from the total generated code amount obtained in ii) above and the total generated code amount obtained this time, and the quantization table RAMs 21 and 22 are generated.
Write to. Further, the total code amount generated by the DC component and the EOB length of each block are subtracted from the target code amount, and the ratio between the subtracted value and the total code amount generated by the AC component is obtained. The AC of each block is added to the above ratio.
The value obtained by subtracting the EOB length from the code amount generated by the component is multiplied, and the value obtained by adding the EOB length to the code amount generated by the DC component of each block is used as the predicted value of the generated code amount of each block. Write to memory 15. vi) and DCT → quantization → Huffman coding (mo
Perform de 2).

[0009]

By the way, in such a conventional image data compression apparatus 10, there is a difference (remainder bit) between the predicted value obtained by the above method and the target generated code amount. Occurs. The reason for the surplus is as follows. In other words, in reality, the above calculation does not completely divide it, but there is always a remainder. When the remainder of this bit is rounded and rounded, an error due to the rounding may cause some bits to be over, and as a result, some bits (or many bytes) may be exceeded as a whole. Since it is not allowed to exceed the target value, the rounding is not actually performed, and the calculation is basically performed by rounding down. When the data is truncated in this way, the number of blocks is 10,000 or more. Therefore, for example, even if each block has an average of 5 bits, the total is 5 blocks.
Thousands of bits will be left. Since the conventional image data compression device 10 has a configuration in which the excess bits are discarded or they are collectively put in the upper left block,
Although it is not possible to improve the image quality by discarding the surplus bits, the image quality is improved by consolidating the blocks in the upper left block over discarding the surplus bits when the generated code amount is smaller than the prediction. However, when the generated code amount is larger than the predicted amount, the code amount allocated to each block is reduced, and thus there is a drawback that the image quality is significantly deteriorated. Therefore, an object of the present invention is to provide an image data compression apparatus capable of appropriately distributing the remainder of the generated code amount and improving the image quality.

[0010]

The invention according to claim 1 is
In order to achieve the above object, transform coding means for transform coding image data, quantization table storage means for storing a quantization table used in a quantization operation, and a signal output from the transform coding means for quantizing Quantizing means for quantizing based on a coding table, coding means for coding the components quantized by the quantizing means in accordance with the predicted generated code amount of each block, and DC component generated in each block at the time of coding. And a code amount storage means for storing the code amount of the AC component, a total generated code amount calculation means for calculating the sum of the code amounts of the DC component and the AD component, and a total generated code calculated by the total generated code amount calculation means. An image data compression apparatus comprising a total generated code amount storage means for storing the code amount, wherein a total of predicted values calculated by the total generated code amount calculation means and a target are set. Determined as the remainder of the difference between the raw code amount,
Each block distributing means for dividing the remainder by the total number of blocks to distribute to each block, Y block distributing means for distributing the remainder to a block of luminance information (Y), and the remainder to luminance information (Y) and color information. Y, RY, BY block distributing means for distributing at a predetermined ratio to the leftmost block of (RY, BY) and Y for distributing the remainder to the leftmost block of luminance information (Y). Of the respective block distributing means of the leftmost block distributing means and the Y upper leftmost block distributing means for distributing the remainder to the upper leftmost block of the luminance information (Y),
At least any one or more block distribution means are provided. According to a second aspect of the present invention, a transform coding means for transform coding the image data, a quantization table storage means for storing a quantization table used in the quantization operation, and a signal output from the transform coding means. Quantizing means for quantizing based on the quantization table, coding means for coding the components quantized by the quantizing means according to the predicted generated code amount of each block, and the coding means generated in each block at the time of coding. Code amount storage means for storing the code amounts of the DC component and AC component, total generated code amount calculation means for calculating the sum of the code amounts of the DC component and AD component, and the total generated code amount calculation means An image data compression device comprising a total generated code amount storage means for storing the total generated code amount, wherein first, a sum of the predicted values calculated by the total generated code amount calculating means and an index The difference between the generated code amounts is calculated as a remainder, the remainder is divided by the total number of blocks and distributed to each block, the remainder is distributed to the blocks of the luminance information (Y), and the remainder is the luminance information (Y). And color information (R
-Y, BY) is distributed to the leftmost block of the luminance information (Y) at a predetermined ratio, the remainder is distributed to the leftmost block of the luminance information (Y), and the remainder is the leftmost block of the luminance information (Y). To be distributed to. The transform coding means may be, for example, a discrete cosine transform operation for performing a discrete cosine transform on image data as described in claim 3, and the coding means may be, for example, As described in 4, the Huffman coding that constitutes a compact code may be performed on an information source given a probability distribution.

[0011]

The operation of the means of the present invention is as follows. According to the first and second aspects of the present invention, the image data is discrete cosine transformed by the transform coding means and then sent to the quantizing means. The quantization means performs a quantization operation on the DCT operation output by referring to the quantization table, and outputs the operation result to the encoding means. In the coding means, the quantized data is coded in accordance with the predicted generated code amount of each block,
At the time of encoding, the DC component and A generated in each block
The code amount of the C component is stored in the code amount storage means. Also,
The total generated code amount calculation means calculates the sum of the code amounts of the DC component and the AC component and stores the sum in the total generated code amount storage means. Then, the difference between the total sum of the calculated prediction values and the target generated code amount is obtained as a remainder, and first, the remainder is distributed to each block, and the remainder remaining after distribution to each block is luminance information (Y). Distributed in blocks. In addition, the remainder is luminance information (Y) and color information (RY, BY).
Of the luminance information (Y) is distributed at a predetermined ratio to the leftmost block of the luminance information (Y). The remaining residue at the end is distributed to the upper left block of the luminance information (Y). Therefore, the remainder of the generated code amount is appropriately distributed, and the image quality can be improved.

[0012]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below with reference to the drawings. 1 to 9 are diagrams showing an embodiment of an image data compression apparatus according to the present invention. In this embodiment, as a method of using a plurality of types of quantization tables, each block has a checkered two types of quantization tables. This is an example of using. First, the configuration will be described. FIG. 1 is a block diagram of the image data compression apparatus 30. In this figure, the image data compression apparatus 30 is
A frame memory 31 for storing original image data to be data-compressed, a DCT device 32 for performing a DCT operation to obtain a DCT coefficient by subjecting the image data read from the frame memory 31 to a two-dimensional DCT, and output data of the DCT operation are quantized. A quantization device 33 for performing a quantization operation according to a value in a quantization table RAM (described later) and two operation modes mode1 and m similar to the conventional example using each of the memories 35 to 39 described later.
The Huffman encoder 34 that performs Huffman coding on the output of the quantizer 33 according to node 2, the DC code amount memory 35 that stores the DC code amount generated by the DC component, and the AC code amount generated by the AC component. Using an AC code amount memory 36 for storing, a DC total generated code amount storage memory 37 for storing the sum of code amounts generated by DC components of all blocks, and two types of quantization tables for Y and C respectively. AC table-based total generated code amount storage memories 38 and 39 for storing the sum total of the code amounts generated by the AC components of the quantized blocks, a quantization table that is considered to be optimal and control of the whole, and a block of each block at that time. It is configured by a CPU 40 that calculates a predicted value of the generated code amount.

The Huffman encoder 34 uses the DC code amount memory 3 for the data output from the quantizer 33.
5 and the AC code amount memory 36, Huffman coding is performed, the sum of the code amounts generated by the DC components of all blocks is stored in the DC total generated code amount storage memory 37, and the checkered pattern (FIG. 3 and FIG. The total sum of the code amounts generated by the AC components of each of the two types of quantization tables 51 to 54, which will be described later, is stored in the table-specific AC total generated code amount storage memories 38 and 39.

Of the table-based AC total generated code amount storage memories 38 and 39, the AC total generated code amount storage memory 38 includes the quantization tables 51 and 52, and the AC total generated code amount storage memory 39. Is the quantization table 53,
The sum of the code amounts generated by the AC components of the block quantized by using 54 is stored and held. At the time of (mode b) described later, the value is written only in the AC total generated code amount storage memory 38, and the value in the AC total generated code amount storage memory 39 is zero.

FIG. 2 is a block diagram of the quantizer 33. In FIG. 2, the quantizing device 33 includes a quantizing table RAM 41 for storing a Y quantizing table 51 in one block of a checkered pattern when each block for C is checkered, and a quantizing table RAM 41 for each C. Quantization table RAM4 for storing the Y quantization table 53 in the other block having a checkered pattern when the block is made checkered.
3 and RY in one block of a checkered pattern when each block for RY and BY is checkered,
A quantization table RAM 42 that stores a BY quantization table 52, and RY and B- in the other block of the checkerboard pattern when each of the RY and BY blocks is checkered. A quantization table RAM 44 that stores the Y quantization table 54, and a multiplication that changes the weighting by multiplying the result of the DCT operation and the values of the quantization tables 51 to 54 (multiplication and division). And a container 45.

FIG. 3 is a diagram showing the Y quantization tables 51 and 53 stored in the quantization table RAMs 41 and 43, and FIG. 54 is a diagram showing the quantization tables 52 and 54 stored in the quantization table RAMs 42 and 54. As shown in FIG. 3, the Y quantization tables 51 and 53 are arranged in a checkered pattern for each block in the Y quantization table 51 and the Y quantization table 53.
(Refer to the hatched portion in FIG. 3). That is, in the conventional example, each block constitutes one Y quantization table as a whole as shown in FIG.
For the quantization tables 51 and 53 for use, each block is in a checkered pattern as shown in FIG. 3, and the quantization table 53 is assigned to the block shown in the hatched portion in FIG. 3 and the quantization table 51 is assigned to the block shown in the non-hatched portion. To do so. Similarly, the RY and BY quantization tables 52 and 54 are in a checkered pattern for each block, and the RY and BY quantization tables 52 and the RY and BY quantization tables are arranged in a checkered pattern. Table 54 (Fig. 4
4), each block is in a checkered pattern as shown in FIG. 4, the quantization table 54 is shown for the blocks shown in the hatched part in FIG. 4, and the quantization table 52 is shown for the blocks shown in the non-hatched part. Try to allocate.

Here, when the checkerboard pattern is actually used, the number of allocated blocks is halved, so the generated code amount is also halved. Therefore, in the present embodiment, the calculated generated code amount is doubled and used. In this way, by switching the quantization tables 51 and 53, 52 and 54, the tendency of the generated code amount at two points (in the case of a linear case, the inclination) can be obtained based on the two tables. The scaling value corresponding to the code amount can be obtained by the first quantization.

Further, a Y quantization table 51 and RY, BY quantization table 52 used as Table 1, and a Y quantization table 53 and RY, BY used as Table 2. For the quantization table for use 54, for example, those having values as shown in FIG. 5 are used.

The quantization tables 51 to 54 stored in the quantization table RAMs 41 to 44 are used by switching the components and blocks as follows. (Mode a) Y quantization table 5 for Y
1,53 are used. Also, this Y quantization table 5
1, 53 are switched for each block in a checkered pattern as shown in FIG. For RY and BY, the RY and BY quantization tables 52 and 54 are used. The switching method is switched in a checkered pattern similar to Y. (Mode b) Y quantization table 51 for Y
, The RY and BY quantization tables 52 are used for RY and BY. In this case, switching for each block is not performed.

Next, the operation of this embodiment will be described. Overall Operation First, it is assumed that the frame memory 31 holds original image data. The DCT calculation unit 32 reads the image data from the frame memory 31 and performs DCT calculation, and the DCT calculation output is sent to the quantizer 33. Further, the quantizing device 33 performs a quantizing operation on the input from the DCT operating device 32 according to the values of the quantization table RAMs 41 to 44, and outputs the operation result to the Huffman encoder 34. The Huffman encoder 34 performs Huffman coding on the quantized data by using the DC code amount memory 35 and the AC code amount memory 36, and D is the sum of the code amounts generated by the DC components of all blocks.
C two kinds of quantization tables 51 to 54 which are stored in the C total generated code amount storage memory 37 and are switched in a checkered pattern
The sum of the code amount generated by each AC component is written in the table-specific AC total generated code amount storage memories 38 and 39. In this case, the sum of the generated code amounts of the blocks using the Y quantization table 51 (see the non-hatched portion in FIG. 3) is written in the AC table-based total generated code amount storage memory 38, and the Y quantization table 51 is used. The total sum of the generated code amounts of the blocks using (see the hatched portion in FIG. 3) is written in the total generated code amount storage memory 39 for each AC table.

The CPU 40 operates each of the above devices as follows to control the generated code amount. i) First, a quantization table is created in accordance with the target total generated code amount, and the quantization tables 51, 52, 53, 5 are created.
Write the reciprocal to 4. FIG. 5 shows an example of the quantization tables 51 to 54 for 1 bpp. ii) DCT → quantization (mode a) → Huffman coding (mode 1) is performed. iii) A quantization table that is considered to be optimal for generating a target code amount according to an algorithm described later, the generated code amount of the AC component and the DC component of each block at that time are predicted, and the value is quantized by the quantization table 51. , 52, the DC code amount memory 35, and the AC code amount memory 36. iv) Next, DCT-> quantization (mode b)-> Huffman coding (mode 2) is performed.

Generation Code Prediction Algorithm First, the quantization table creation algorithm will be described. A method of creating a quantization table that makes the generated code amount linear is determined and numbered in advance. This number will be called scaling. For example, it is assumed that there are 256 ways from 0 to 255, and the generated code amount increases linearly from 0 to 255. Two appropriate scalings are selected for the target generated code amount (CDN), a quantization table is created, the reciprocal numbers are written in the quantization tables 51, 52, 53, 54, and quantization (mode a) is performed. And Huffman coding (mo
Perform de 1). FIG. 5 is an example of the quantization table created in this way. The scaling values selected first are Sb and Sh.
Further, the value of the total DC generated code amount in the total DC generated code amount storage memory 37 is DCall, and the AC corresponding to the above Sb and Sh.
The values of the AC total generated code amount for each table in the table total generated code amount storage memories 38 and 39 are respectively ACsb and ACs.
If h is set, the required scaling can be obtained by, for example, Equation 1.

[Equation 1]

Next, the generated code amount of each block is predicted by the following algorithm. 6 to 8 are flowcharts showing the prediction processing of the generated code amount of each block, and the correspondence of the variables used in the flowcharts of FIGS. 6 to 8 is shown below. CDN: Target number of bits RN: Unallocated number of bits RAT [0], RAT [1]: Generated code amount of AC component in target number of bits, and quantization tables 51, 52
16 times the ratio predicted value with double the total generated code amount when 53 and 54 are used. EOB [0], EOB [1]: 0: Y EOB length 1: C EOB length AC [i] [J] [k]: i → 0/1 = Y / C j = k, AC generated code amount of the block at row j and column DC [i] [j] [k]: i → 0/1 = Y / Cj = k, DC generated code amount of block of j row and k column F [i] [j] [k]: i → 0/1 = Y / C j = k, generated code amount of block of j row and k column Prediction value RA: Number of bits distributed to all blocks YA: Number of bits distributed to Y blocks RB: Number of bits distributed to all left blocks YB: Number of bits distributed to Y left blocks

In explaining the flowcharts of FIGS. 6 to 8, first, the algorithm of the entire flow will be briefly described. (I) D generated at the first time from the target generated code amount
The ratio of the value obtained by subtracting the code amount of the C component and the total sum (ACsb, ACsh) of the AC components generated by each table is obtained. (II) From the AC generated code amount actually generated in each block, the EOB length (EOB [0],
EOB [1]) is subtracted, the ratio obtained in (I) above is multiplied according to the quantization table used at one time, and the value obtained by adding EOB and the DC generated code amount obtained at the first time to a temporary value Use as a predicted value. (III) The difference between the sum of the prediction values obtained in (II) above and the target generated code amount is obtained, and each block is allocated as follows. a) Divide the remainder by the total number of blocks and distribute evenly to each block. b) Furthermore, the remainder is divided by the number of Y blocks, and is evenly distributed only to the Y blocks. c) Further, the remainder is divided by 4 times the number of blocks in the V direction, and distributed to the leftmost block of Y, RY and BY at a ratio of 2: 1: 1. d) Evenly distribute the remainder to the leftmost block in Y. e) Furthermore, the remainder is added to the upper left block of Y. The predicted generated code amount of each block is obtained by the above algorithm.

More specifically, in FIGS. 6 to 8, first, in step S1, the unallocated number of bits RN is set as the target number of bits CDN (RN = CDN), and the target number of bits CDN is first calculated. Ratios RAT [0] and RAT [1] of a value obtained by subtracting the total generated code amount DC all of the generated DC components and the total sum (AC sb, AC sh) of the AC components generated by each table are calculated (RAT
[0] = (CDN-DCall) × 8 / AC ch, R
AT [1] = (CDN-DC all) × 8 / AC c
h). Here, the value (CDN-DC all) is set to x8 in order to shift the binary point to maintain the calculation accuracy and is not necessarily essential. For example, if you create a compression / decompression program in C,
Since the decimal part of language division is truncated, multiply it by an appropriate multiple here.

Next, in step S2, a bank i indicating a Y or C block is set to 0 (i =
0), whether the bank i is smaller than 2 in step S3 (i <
2) or not. That is, since the color (C) is sampled twice in the horizontal direction, the two of bank 1 and bank 2 have the same size as Y. for that reason,
The subsequent processing is repeated until the bank i becomes 2 using the arrays in the banks 1 and 2. FIG. 9 is a diagram showing the structure of Y, BY, and RY blocks, and each block is shown in j rows and k columns. When bank i is smaller than 2, step S
In step 4, the number j of blocks in the vertical direction is set to 0 (j = 0), and in step S5, it is determined whether or not the number j of blocks in the vertical direction is smaller than a predetermined number V of blocks in the vertical direction (j <V). When the number j of blocks in the vertical direction is equal to or larger than the predetermined number V of blocks in the vertical direction, the bank i is incremented (i = i +) in step S6.
0) and returns to step S3. When the number j of blocks in the vertical direction is smaller than V, the number k of blocks in the horizontal direction is set to 0 in step S7 (k = 0), and the number k of blocks in the horizontal direction is smaller than the predetermined number H of blocks in the horizontal direction in step S8. (K <
H)) or not. When the horizontal block number k is equal to or larger than the predetermined horizontal block number H, the vertical block number j is incremented (j = j + 1) in step S9, and the process returns to step S5. For example, the number V of blocks in the vertical direction is 6
If it is 0, the number of blocks j in the vertical direction is initialized to 0. Therefore, when counting from j = 0 to j = 60, the same determination is made in the horizontal direction at step S7 and thereafter. As a result, the process shown in the next step S10 is repeatedly executed until the number of blocks specified in the vertical and horizontal directions is reached, and when the bank i becomes 2, the flow of FIG. 6 is skipped and the process proceeds to step S11 of the flow of FIG.

In step S10, the following processes are performed. First, the EOB length EOB [i] corresponding to each component is subtracted from the AC generated code amount AC [i] [j] [k] actually generated in each block and stored in the temporary temp (temp = AC [i]). [J] [k] -EOB
[I]). Here, EOB [i] is subtracted from the AC generated code amount AC [i] [j] [k] generated in each block because EOB is added in order to prevent overflow in later processing. This EO will be done in advance.
The B length is subtracted. Further, a value obtained by multiplying the temporary temp by a predetermined standing RAT is set as a new temporary temp value (temp = temp * RAT [(j + k)).
% 2]). Here, (j + k)% 2 is for switching 0 and 1 in a checkered pattern (see FIGS. 3 and 4) in each block, and the above% 2 is the remainder divided by 2 (0 or 1). Either will be shown). This determines which checkered ratio RAT to use. Further, the value of this temporary temp is shifted to the left by 4 bits (temp = (temp >> 4). Here, shifting to the left by 4 bits is to restore the value multiplied by × 8 in step S1 to the original value. A DC temporary code amount DC [i] [j] [k] obtained at the first time and EOB [i] corresponding to the component are added to the value of the temporary temp to obtain a new temporary te.
As the value of mp (temp = DC [i] [j] [k] + EOB
[I]), and the value of the temporary temp is defined as the generated code amount F [i] [j] [k] of the block (F [i] [j]).
[K] = temp). This F [i] [j] [k] becomes the actual predicted code amount. Further, the unallocated bit number (remaining bit number) RN is obtained by subtracting the value of the temporary temp obtained above from the previously unallocated bit number RN (RN = RN-temp). Then, the number of blocks in the horizontal direction k
Is incremented (k = k + 1) to end the process of step S10 and return to step S8. The predicted generated code amount F [i] [j] [k] of each block can be obtained by the flow of FIG.

By the way, there is a case where the algorithm shown in FIG. 6 is sufficient for a device having a very high calculation accuracy (for example, an image data compression device using an FPU). The surplus portion is allocated by the algorithm as shown in the flow of (1) to obtain a more accurate predicted code amount, and the image quality information is greatly improved.

In FIG. 7, in step S11, the following processes are performed. First, the unallocated remaining bit number R
N is divided by the total number of blocks (V × H × 2) to obtain the number of bits RA to be distributed to all blocks (RA = RN /
(V × H × 2)). For example, assuming that the Y block shown in FIG. 9 has 100 bits, the BY and RY blocks each have 50 bits, and the number of remaining bits is 380, the remainder 380 is represented by Equation 2 as a whole block number ( 100 + 50 + 5
The number of bits RA to be distributed to each block, for example, by 1 bit, is obtained by performing division by dividing by 0). 380 / (100 + 50 + 50) = 1 = RA (Equation 2) Then, the number of bits (RA × V × H × 2) to be distributed to each block is subtracted from the original unallocated number of bits RN, and distribution is performed by the above-mentioned equal distribution. A surplus number of bits RN that cannot be completely obtained is obtained (RN = RN−RA × V × H × 2).

Further, the surplus number of bits RN that could not be distributed by the above-mentioned even distribution is divided by the number of Y blocks (V × H) to obtain the number of bits YA which is evenly distributed only to the Y blocks (YA = RN). / (VxH)). That is, the image quality becomes better when Y and color (C) do not have the same weight and more codes are assigned to Y. Therefore,
The surplus number of bits RN is distributed only to Y blocks. In the above example, YA = RN / (V × H) = 180 /
It becomes 100. Then, even if the number of bits (YA × V × H) to be distributed only to the Y block is subtracted from the surplus number of bits RN to be uniformly distributed to each block and evenly distributed to only the Y block, the distribution cannot be completed. Remaining number of bits R
Find N (RN = RN-YAV * H). In the above example, RN = 180-100.

As described above, the blocks are evenly distributed only to the Y blocks, but there is still a case that the remainder still occurs. Therefore, the surplus bit number RN that could not be distributed by the above Y distribution is divided by 4 times the number of blocks in the V direction, and Y,
The number of bits RB to be distributed to the leftmost block of RY and BY is calculated at a ratio of 2: 1: 1 (RB = RN / (V
× 4). That is, normally, processing is performed for blocks in the horizontal direction from the upper left block, but appropriate distribution can no longer be performed even if there is a large amount of excess when it comes to the lower right block. Therefore, in this embodiment, the blocks are distributed to the leftmost block. Also in this case, since Y is given a larger number of codes, for example, Y is given twice as much as RY and BY, and the leftmost block of Y, RY, and BY has 2 blocks, respectively. Distribute at a ratio of 1: 1. In order to realize this, the above RN is divided by (V × 4). In the above example, if V = 10, RB = 80 / (10 × 2 + 1
0 + 10) = 2. Then, even if the number of bits (RB × 4 × 4) to be distributed to all left blocks is subtracted from the surplus number of bits RN to be distributed in the Y block and distribution is made to the leftmost block, the distribution cannot be completed. The number of remaining bits RN is obtained (RN = RN-RB * V * 4).

Even if the above-mentioned distribution of 2: 1: 1 to the leftmost block of Y, RY, and BY, the remainder may still occur. Therefore, the remaining bit number RN that could not be distributed is divided by the number of blocks in the V direction,
Number of bits YB evenly distributed to the leftmost block of Y
Is calculated (YB = RN / V). And Y, RY, B
-The number of surplus bits R distributed to the leftmost block of Y
Number of bits to distribute to the leftmost block from N to Y (Y
(B × V) is subtracted and evenly distributed to the leftmost block of Y, the remaining bit number RN that cannot be distributed is obtained (RN = RN−YB × V). If there is still a remainder even if it is evenly distributed to the leftmost block of Y described above, the remainder is added to the block at the upper left end of Y in step S15 described later.

With the above, the processing of step S11 is completed,
Go to step S12. In step S12, the number of blocks in the vertical direction j is set to 0 (j = 0) in order to perform processing in the vertical direction, and in step S13, the number of blocks in the vertical direction j
Is smaller than the predetermined number of blocks V in the vertical direction (j <V)
Determine whether or not. If the number j of blocks in the vertical direction is greater than or equal to the predetermined number V of blocks in the vertical direction, and in step S14 if the number of blocks in the vertical direction is greater than or equal to the predetermined block number V, the generated code amount prediction value F [i] [j] [k] is obtained. The number of blocks j in the vertical direction is incremented (j = j + 1) and the process returns to step S13. Here, the predicted generated code amount F [i] [j] [k]
Among them, [i] is a bank, and Y is i when i = 0, and BY and RY when i = 1. Also,
[I] is the V (vertical) direction, and [k] is the H (horizontal) direction.
In this case, the predicted generation code amount F [0] [j] of Y
[0] is the predicted generated code amount F [0] obtained at the first time.
[J] [0] is a value obtained by adding a value obtained by doubling the number of bits RB to be distributed to all the leftmost blocks as a remainder and the number of bits to be distributed to the leftmost block of Y (F [0]
[J] [0] = F [0] [j] [0] + RB × 2 + Y
B). In addition, the prediction generated code amount F [1] [j] of BY is used.
[0] is the predicted generated code amount F [1] obtained at the first time.
It becomes a value obtained by adding the number of bits RB to be distributed to all the leftmost blocks as a remainder to [j] [0] (F [1]
[J] [1] = F [1] [j] [0] + RB), and
The predicted generated code amount F [1] [j] [H / 2] of RY is
Predicted code amount F [1] [j] [H /
2] is added with the number of bits RB to be distributed to all leftmost blocks as a remainder (F [1] [j]
[0] = F [1] [j] [H / 2]). Here, as shown in FIG. 9, two BY and BY are combined to be Y, so that the horizontal [k] of the predicted generated code amount of RY is [H / 2].
It is represented by. In this way, the predicted generated code amount F [i] [j] [k] in the loop of steps S13 and S14 described above.
Is repeatedly calculated until V becomes the number of blocks in the vertical direction,
The remainder will be allotted to the leftmost block.
When the number j of blocks in the vertical direction becomes equal to or larger than the predetermined number V of blocks in the vertical direction, it is determined that processing of all blocks in the vertical direction has been completed, and in step S15, the remainder RN at the final stage is set to the prediction generation code amount F of the leftmost block. In addition to [0] [0] [0], (F [0] [0] [0] = F [0] [0] [0] + R
N) Proceed to step S16 in FIG.

In the flow shown in FIG. 8, the above-mentioned remainder is equally distributed to all blocks and the Y-block is equally distributed. That is, step S1
In step 6, the vertical block number j is set to 0 (j = 0), and in step S17, the vertical block number j is smaller than V (j
<V)) or not. If the number j of blocks in the vertical direction is smaller than the predetermined number V of blocks in the vertical direction, step S18.
The number k of horizontal blocks is set to 0 (k = 0), and it is determined in step S20 whether the number k of horizontal blocks is smaller than a predetermined number H of horizontal blocks (k <H). When the number k of blocks in the horizontal direction is equal to or greater than the predetermined number H of blocks in the horizontal direction, the number j of blocks in the vertical direction is incremented (j = j + 1) in step S19, and the process returns to step S17. If it is smaller, the process proceeds to step S21.

In step S21, the predicted generated code amount F [0] [j] [k] of Y which has already been calculated in the preceding process.
And the number of bits RA to be distributed to all blocks and the number of bits YA to be distributed to blocks of Y are added to obtain the predicted generation code amount F [0] [j] [k] of Y and (F [0]
[J] [k] = F [0] [j] [k] + RA + YA),
The number of bits RA to be distributed to all blocks is added to the prediction generated code amount F [1] [j] [k] of C already calculated in the previous step, and the number of bits YA to be distributed to the block of C is added. Then, the predicted generated code amount F [1] [j] of Y
[K] is calculated (F [1] [j] [k] = F [1] [j]
[K] + RA + YA), the horizontal block number k is incremented (k = k + 1), and the process returns to step S20.

Each of the above steps is repeatedly processed, and when the number of blocks j in the vertical direction becomes V or more in step S17, Y,
When it is determined that the predicted generated code amount has been calculated for all blocks of C, the process of this flow ends.

As described above, the image data compressing apparatus 30 obtains the difference between the sum of the calculated predicted values and the target generated code amount, divides the remainder by the total number of blocks, and evenly distributes each block. And further divide the remainder by the number of Y blocks to evenly distribute only to the Y blocks, and
Divide the remainder by 4 times the number of blocks in the V direction,
2: 1: 1 in the leftmost block of Y, RY, BY
And the remainder is evenly distributed to the leftmost block of Y, and the remainder is added to the block at the upper left corner of Y to obtain the predicted generated code amount of each block. Therefore, it is possible to prevent the problem that the image quality is significantly deteriorated when the generated code amount is larger than the prediction, and it is possible to appropriately distribute the surplus to enhance the image quality.

Further, in this embodiment, when the first compression is performed, 2
By using one of the quantization tables 51 and 53 and 52 and 54, the change in the generated code amount due to the change of the quantization table is predicted, and the optimum quantization table for the target generated code amount is created. Since it is designed to compress to the target code amount by compression, it is possible to do the operation that was previously performed with the "DCT → quantization → encoding" sequence of 3 times in 2 operations. The processing speed can be significantly increased. Although the remainder is distributed by the algorithm shown in FIGS. 6 to 8 in this embodiment, it is not always necessary to perform all the distribution methods described in this embodiment in the disclosed order. .. For example, after evenly distributing the remainder to each block,
A mode in which the remainder is simply added to the leftmost block of Y may be used.

Although a plurality of types of quantization tables are used in the present embodiment, it goes without saying that a plurality of types of quantization tables are used as long as the remainder is allocated.

In the present embodiment, the image data compression apparatus is applied to the image data compression apparatus based on the JPEG algorithm, but the present invention is not limited to this, and any apparatus can be used as long as it controls the generated code amount. Needless to say, it can be applied to the above device. Further, any method may be used as long as it uses a quantization table, and the method of creating the quantization table is not limited to that in this embodiment.

In the present embodiment, the Huffman coding method is used in the coding unit. However, the present invention is not limited to this. For example, an arithmetic code may be used as entropy coding instead of a block code such as Huffman code. Alternatively, Q-Coder, which is a type of arithmetic code, may be used.

Further, in this embodiment, the conversion coding method is set to D.
Although CT is applied, the present invention is not limited to this DCT method, and is applied to, for example, an image data compression device using Hadamard transform, Harr transform, gradient transform (slant transform), symmetric sine transform, or the like. be able to.

Further, it goes without saying that the numbers and types of circuits and members constituting the image data compression device 30 and the quantization device 33 are not limited to those in the above-mentioned embodiment.

[0044]

According to the invention described in claims 1, 2, 3 and 4, each block distributing means distributes the remainder of the difference between the sum of the calculated predicted values and the target generated code amount to each block. , Y for distributing the remainder to blocks of luminance information (Y)
Block distribution means, and Y, RY, BY block distribution means for distributing the remainder to the leftmost block of luminance information (Y) and color information (RY, BY) at a predetermined ratio. Y to distribute the remainder to the leftmost block of luminance information (Y)
At least one of the leftmost block distributing means and the left uppermost block distributing means for distributing the remainder to the upper leftmost block of the luminance information (Y).
Since one or more block distributing means are provided, the remainder of the generated code amount can be appropriately distributed, and the image quality in the code amount control can be improved.

[Brief description of drawings]

FIG. 1 is a block configuration diagram of an image data compression device.

FIG. 2 is a block configuration diagram of a quantization device of an image data compression device.

FIG. 3 is a diagram showing a Y quantization table of the image data compression apparatus.

FIG. 4 is a diagram showing RY and BY tables of the image data compression apparatus.

FIG. 5 is an example of a quantization table of the image data compression device.

FIG. 6 is a flowchart showing a prediction process of a generated code amount of the image data compression device.

FIG. 7 is a flowchart showing a prediction process of a generated code amount of the image data compression device.

FIG. 8 is a flowchart showing a prediction process of a generated code amount of the image data compression device.

FIG. 9 is a diagram showing a structure of Y, BY, and RY blocks of the image data compression device.

FIG. 10 is a block diagram of a conventional image data compression device.

FIG. 11 is a block configuration diagram of a quantization device of a conventional image data compression device.

FIG. 12 is a diagram showing a Y quantization table of a conventional image data compression apparatus.

FIG. 13 shows RY and BY of a conventional image data compression device.
It is a figure which shows the quantization table for.

[Explanation of symbols]

 30 image data compression device 31 frame memory 32 DCT operation device 33 quantizer 34 Huffman encoder 35 DC code amount memory 36 AC code amount memory 37 DC total generated code amount storage memory 38, 39 total generated code amount by AC table Memory 40 CPU 41-44 Quantization table RAM 45 Multiplier 51-54 Quantization table

Claims (4)

[Claims] 1. A transform coding means for transform coding image data, a quantization table storage means for storing a quantization table used in a quantization operation, and a signal output from the transform coding means for quantizing the signal. Quantizing means for quantizing based on a table, coding means for coding the components quantized by the quantizing means according to the predicted generated code amount of each block, DC component generated in each block at the time of coding, and A
A code amount storage unit that stores the code amount of the C component, a total generated code amount calculation unit that calculates the sum of the code amounts of the DC component and the AD component, and a total generated code amount calculated by the total generated code amount calculation unit. An image data compression apparatus comprising a total generated code amount storage means for storing an amount, wherein a difference between a total sum of predicted values calculated by the total generated code amount calculation means and a target generated code amount is obtained as a remainder. , Each block distributing means for dividing the remainder by the total number of blocks and distributing it to each block, Y block distributing means for distributing the remainder to blocks of luminance information (Y), and the remainder for luminance information (Y) and color Information (RY, B-
Y), which is distributed to the leftmost block of Y) at a predetermined ratio,
Y, BY block distributing means, Y left end block distributing means for distributing the remainder to the leftmost block of luminance information (Y), and Y upper left for distributing the remainder to the upper left block of luminance information (Y) An image data compression apparatus comprising at least one or more block distributing means among the end block distributing means and the block distributing means. 2. A transform coding means for transform coding image data, a quantization table storage means for storing a quantization table used for a quantization operation, and a signal output from the transform coding means for the quantization. Quantizing means for quantizing based on a table, coding means for coding the components quantized by the quantizing means according to the predicted generated code amount of each block, DC component generated in each block at the time of coding, and A
A code amount storage unit that stores the code amount of the C component, a total generated code amount calculation unit that calculates the sum of the code amounts of the DC component and the AD component, and a total generated code amount calculated by the total generated code amount calculation unit. An image data compression apparatus comprising a total generated code amount storage means for storing an amount, wherein a difference between a total sum of predicted values calculated by the total generated code amount calculation means and a target generated code amount is left over. Then, the remainder is divided by the total number of blocks and distributed to each block, the remainder is distributed to blocks of luminance information (Y), and the remainder is further divided into luminance information (Y) and color information (RY, B).
-Y) is distributed to the left end block of the luminance information (Y) at a predetermined ratio, and the remainder is distributed to the left end block of the luminance information (Y),
Further, the image data compression method is characterized in that the remainder is distributed to the upper left block of the luminance information (Y). 3. The image data compression method according to claim 1, wherein the transform coding means is a discrete cosine transform operation for performing a discrete cosine transform on the image data. apparatus. 4. The encoding device according to claim 1, wherein the encoding means performs Huffman encoding forming a compact code on an information source given a probability distribution. Image data compression apparatus and image data compression method.