JPH10136360A - Image data compression system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an image data compression system by which fixed length processing is conducted at high speed with high accuracy in the case of compressing a digital image to reduce a data amount. SOLUTION: This system is provided with a DCT means 2 that applies discrete cosine transformation(DCT) to image data in the unit of blocks to generate a DCT coefficient, code data generating means 4, 5 that generate code data at 1st and 2nd data compression degrees as to the DCT coefficient every time the DCT coefficient of one block is generated, and a data compression degree estimate means 7 that accumulates amount of code data of the 1st and 2nd data compression degrees generated by the code data generating means 4, 5 as to all blocks to be received and estimates a 3rd data compression degree and a 4th data compression degree higher than the 3rd data compression degree to generate code data with an object data amount in response to the code data amount of both the 1st and 2nd data compression degrees.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a digital image processing system, and more particularly to an image data compression system capable of compressing digital image data to reduce the amount of data.

[0002]

2. Description of the Related Art A digital still camera is one of those using an image data compression system. A digital still camera shoots a digital still image by pointing a lens at a subject and pressing a shutter button. The image formed via the lens is converted into an electric signal, compressed, and stored in a replaceable memory card or the like. Data compression is a process for reducing the amount of data and storing a large amount of image data in a memory card.

The amount of code data obtained by compressing digital image data differs depending on the spatial frequency distribution of the digital image. For example, for a digital image containing many high-frequency components, the amount of code data cannot be reduced much. On the other hand, with respect to a digital image having few high frequency components, the amount of code data can be considerably reduced. That is, the amount of code data generated by data compression generally differs depending on the type of digital image, although it differs depending on the data compression method.

[0004] Code data obtained by data compression is stored in a storage medium such as a memory card. The memory card has a storage capacity of, for example, 1 Mbyte, and in that case, data of 1 Mbyte or more cannot be stored.

[0005] It is desirable to notify the photographer of the remaining recordable number in order to prevent writing of code data exceeding 1 Mbyte in the memory card or for the convenience of the photographer. If the code data to be data-compressed has the same data amount for each image regardless of the type of digital image, the number of digital images that can be recorded on the memory card can be easily notified to the photographer.

However, when the amount of code data is variable, it is impossible to notify the photographer of the remaining number. If the code data amount of the image to be captured is small, a large number of images can be recorded, and if the code data amount of the image to be captured is large, only a small number can be recorded.

Therefore, it is desired to perform a fixed length process of the code data when compressing the digital image data. By performing the fixed length processing, any kind of digital image can be converted into a substantially constant amount of code data. Fixed length processing is one piece (one frame)
This is a process for compressing the digital image data and generating fixed-length code data. If the code data has a fixed length, the photographer can be notified of the remaining number.

Next, the fixed length processing will be described. To perform the fixed-length processing, first, statistical processing is performed as preprocessing, and the data compression degree of data compression is adjusted according to the result of the statistical processing to generate fixed-length code data.

When the photographer presses the shutter button, a digital image is captured. Next, statistical processing is performed on the captured digital image. Statistical processing is processing for estimating how much code data is generated when data compression is performed on captured digital image data.

When the statistical processing is completed, data compression processing and storage processing are performed. As a result of the statistical processing, if it is estimated that a large amount of code data is likely to be generated, the data compression may be performed by setting a high data compression degree. If it is estimated that code data is likely to be generated in a small amount, compression may be performed by setting the data compression degree to a low level. Code data generated by data compression always has a substantially constant data amount.

[0011] After that, the code data that has been subjected to the data compression by the storage process is recorded on the memory card. Above,
A series of processes from capture of the digital image to recording on the memory card ends.

[0012]

In the image compression processing, there is the following method for performing the fixed length processing. First, as statistical processing, data compression processing is performed once at a reference data compression degree. As a result, when a data amount larger than the target data amount is generated, a data compression degree higher than the reference data compression degree is set. On the other hand, when the amount of code data smaller than the target data amount is generated, a data compression degree lower than the reference data compression degree is set. Formal data compression processing is performed at the set data compression degree, and code data of an image is generated.

However, since the relationship between the degree of data compression and the amount of code data is not constant, there is a difference between the target data amount and the generated code data amount depending on the type of image. The accuracy of the fixed length is low only by performing the statistical processing once.

There is a method of performing a statistical process two or more times to determine the data compression degree in order to achieve a more precise fixed length. The case where the statistical processing is performed twice will be described as an example. First, 1
As the third statistical processing, the compression processing is performed at the first reference data compression degree, and the code data amount is estimated. Next, as a second statistical process, a data compression process is performed at a second reference data compression degree different from the first reference data compression degree, and the code data amount is estimated. A formal data compression degree is set in consideration of the above two code data amounts. For example, when the target data amount is between the two code data amounts, a formal data compression degree may be set between the first reference data compression degree and the second reference data compression degree. Formal data compression processing is performed at the set data compression degree, and code data of an image is generated.

However, in this method, since statistical processing is performed twice and formal data compression processing is performed once, a total of three times of data compression processing are performed, which requires a large amount of time.

An object of the present invention is to provide an image data compression system that performs high-speed and high-precision fixed length processing.

[0017]

An image data compression system according to the present invention comprises an image data supply means for supplying image data by dividing the image data into blocks, and an image data supply means for supplying the image data in blocks supplied from the image data supply means. DCT means for generating DCT coefficients for each block by performing a discrete cosine transform (DCT);
Each time a CT coefficient is generated, a first
And code data generating means for generating code data with the second data compression degree, and the amount of code data for each of the first and second data compression degrees generated by the code data generation means. A counter that accumulates for the block of
A third data compression degree considered to be appropriate for generating code data of a target data amount, and a third data compression degree higher than the third data compression degree, in accordance with both the code data amounts of the first and second data compression degrees Data compression degree estimating means (7) for estimating a fourth data compression degree of the data compression degree.

When the DCT means has finished generating the DCT coefficients of the first block, the code data generating means executes the DCT coefficient processing.
For the coefficients, code data is generated at a first data compression degree, and subsequently, code data is generated at a second data compression degree.
After generating the DCT coefficients of the first block, the DCT means can generate the DCT coefficients of the second block. When the DCT means has finished generating the DCT coefficients of the first block, the code data generating means generates code data with at least two types of data compression, and the DCT means generates the DCT coefficients of the second block. Processes can be parallelized. By performing the steps in parallel, the speed of the compression processing can be increased. Further, the data compression degree estimating means can estimate a third data compression degree for generating the target code data amount and a fourth data compression degree higher than the third data compression degree. . If the third data compression degree is appropriate, formal compression may be performed using the third data compression degree. When code data larger than the target code data amount is generated at the third data compression degree, a smaller code data can be generated by using the fourth data compression degree.

[0019]

FIG. 2 is a flowchart showing a processing procedure performed by an image data compression system according to an embodiment of the present invention.

In step S1, statistical processing is performed. This statistical processing calculates the amount of code data for two types of data compression during one compression time. Usually, to calculate the amount of code data for two types of data compression,
At least two data compression times are required. The feature of the statistical processing here is that code data for two types of data compression is efficiently generated and processed in a short time. After generating the code data, two formal data compression degrees are determined based on the amounts of the two types of code data.

The two types of data compression are a first candidate data compression and a second candidate data compression. The data compression degree of the first candidate is a data compression degree calculated as an optimum value. The data compression degree of the second candidate is used when the data compression degree of the first candidate is inappropriate.

In step S2, formal data compression processing is performed using the data compression degrees of the first and second candidates determined by the statistical processing. By this data compression processing, two types of code data of an image are generated. The amount of code data generated based on the data compression degree of the first candidate can approach the target data amount with high accuracy.

Since the statistical processing estimates the approximate code data amount, even if data compression is performed at the first candidate data compression degree, the same code data amount as the target code data amount is generated. Not exclusively. In some cases, the generated code data amount is larger or smaller than the target code data amount.

The code data is finally stored in a storage medium such as a memory card. The storage medium has a finite storage capacity, and cannot store code data exceeding the maximum storable capacity. In the image data compression system, care must be taken not to generate a code data amount larger than the target code data amount.

Therefore, as a result of performing data compression at the data compression degree of the first candidate, if a code data amount larger than the target code data amount is generated, the data is generated based on the data compression degree of the second candidate. Use coded data. The data compression degree of the second candidate is higher than the data compression degree of the first candidate. According to the data compression degree of the second candidate, a small amount of code data can be generated.

If code data larger than the target code data amount is generated even by the data compression degree of the second candidate, a new data compression degree is determined with reference to the data compression degrees of the first and second candidates. I do. Then, data compression is performed again using the new data compression degree. By performing data compression again, it is possible to generate code data having a data amount closer to the target code data amount.

Thus, the fixed-length data compression processing ends. By calculating the data compression degrees of the first and second candidates and performing data compression using the data compression degrees, high-precision fixed-length compression can be performed.

FIG. 1 is a block diagram showing the configuration of an image data compression system according to this embodiment. This image data compression system uses a JPEG (joint photographic expert group) which is a standard compression method for digital still images.
Generate code data of the system. The resources of the conventional JPEG system can be used as they are.

The image data compression system includes an image memory 1, a discrete cosine transform (hereinafter referred to as DCT) unit 2, D
It has a CT coefficient memory 3, a quantization unit 4, an encoding unit 5, a code data amount counting unit 6, a scale factor determination unit 7, and a controller 8. The controller 8 exchanges timing signals with all other processing blocks and adjusts timing between the processing blocks.

Next, each processing block will be described.
The image memory 1 is, for example, a DRAM or a flash memory, and can store one frame of image data. The image memory 1 usually stores image data in a raster format. The image data is composed of a plurality of pixel data.

The raster format is an arrangement of the following pixel data for one frame image. First, the images are arranged sequentially starting from the pixel at the upper left corner of the image toward the right horizontal direction. After reaching the rightmost pixel, subsequently, starting from the leftmost pixel of the next line, the pixels are sequentially arranged in the right horizontal direction. Hereinafter, the process is similarly performed up to the bottom line. The pixel at the lower right corner is the last data.

The image data compression system basically has an 8
Since processing is performed for each block of × 8 pixels, the image memory 1
Converts the image data from the raster format to the block format and supplies the data to the DCT unit 2. A monochrome image has one type of image data. A color image is divided into luminance data and color data.
Perform block conversion.

The block format is an arrangement of the following pixel data for one frame image. One frame image is divided into a plurality of blocks. One block is 8 × 8 pixels. The order of the blocks in one frame starts from the block at the upper left corner and is arranged in the right horizontal direction as in the above-mentioned raster format. The last block is the block in the lower right corner. The arrangement of the pixel data in the block is also the same as in the raster format, starting from the pixel data at the upper left corner in the block and lining up in the right horizontal direction. The last pixel data is the pixel data at the lower right corner in the block.

The DCT unit 2 is supplied with image data I in block format. Hereinafter, processing is performed with one block of image data as one unit. In other words, JPEG compression is 1
The image is divided into blocks of 8 × 8 pixels, and the following processing is performed for each block.

The DCT unit 2 performs a DCT process on the image data I in block units. In the DCT processing, the image data I is interposed between a transposed cosine coefficient matrix ^Dt and a cosine coefficient matrix D, and a matrix operation is performed to obtain a DCT coefficient F.

F = D ^t ID Here, the DCT coefficient F is an 8 × 8 matrix and indicates a spatial frequency component.

The DCT coefficient memory 3 is, for example, a DRAM or an SRAM, and has a DCT coefficient F generated by the DCT unit 2.
Is stored.

Next, the configuration of the quantization unit 4 will be described. The quantization table 4 includes a reference quantization table memory 11, a multiplier 12, a multiplexer 71, and a divider 15. The memory 11 stores a reference quantization table Q.

FIG. 3 shows an example of the reference quantization table Q. As described above, the image data compression system is 8 × 8
, The quantization table Q is configured by an 8 × 8 matrix.

The reference quantization table Q is a quantization table for performing data compression at a standard data compression degree.
The quantization process divides the 8 × 8 DCT coefficient F by the corresponding coefficient in the quantization table Q. The DCT coefficients have lower spatial frequency components in the upper left direction of the matrix, and higher frequency components in the lower right direction. The reference quantization table Q is
As a whole, a lower frequency component performs finer quantization and a higher frequency component performs coarser quantization. Generally, data compression is performed by removing information on high-frequency components of image data in consideration of human visual characteristics and high-frequency components having a lot of noise.

In FIG. 1, a multiplexer 71 controls a first scale factor SF under the control of a controller 8.
a or the second scale factor SFb is supplied to the multiplier 12. The multiplier 12 multiplies the reference quantization table Q by a scale factor SFa or SFb, and outputs a first quantization table Q × SFa or a second quantization table Q × SFb.

The scale factor SF differs between the first quantization table and the second quantization table. The data compression degree of the code data is determined by the value of the scale factor SF. That is, a difference between the first quantization table and the second quantization table indicates a difference in data compression degree.

When the multiplexer 71 supplies the first scale factor SFa to the multiplier 12, the divider 15
The following operation is performed to output the quantized data Ruv. The quantized data Ruv is data that is quantized by the first quantization table. Rounding round means rounding to the nearest integer, and the DCT coefficient Fuv is D
The coefficients are stored in the CT coefficient memory 3.

Ruv = round [Fuv / (SFa × Quv)] On the other hand, the multiplexer 71 sets the second scale factor S
When Fb is supplied to the multiplier 12, the divider 15 performs the following operation and outputs quantized data Ruv. The quantized data Ruv is data that is quantized by the second quantization table.

Ruv = round [Fuv / (SFb × Quv)] The encoding unit 5 performs an encoding process on the quantized data Ruv. The encoding process includes processes of run-length encoding and Huffman encoding. The run-length encoding can perform high compression on data in which a value of 0 continuously follows. In the quantized data Ruv, many zeros tend to be collected in the lower right part (high-frequency component) of the matrix. By utilizing this property and performing run-length encoding of the matrix Ruv of quantized data by zigzag scanning, high compression can be performed. Zigzag scanning is a method of sequentially scanning from low frequency components to high frequency components.

The encoding unit 5 performs Huffman encoding after performing run-length encoding, and generates encoded data.

The code data amount counting section 6 has a counter A
And a counter B. The counter A counts a code data amount NA when the code data is generated by performing quantization using the first quantization table (scale factor SFa). The counter B counts a code data amount NB when code data is generated by performing quantization using the second quantization table (scale factor SFb).

In other words, in the statistical processing, the counter A counts the code data amount NA when compressed at the first reference data compression degree (scale factor SFa). The counter B has a code data amount NB when compressed at the second reference data compression degree (scale factor SFb).
Count.

FIG. 4 shows a method in which the code data amount counting section 6 counts the code data amount. One frame image is composed of, for example, n blocks. Code data is generated in block units. The counters A and B accumulate the amount of code data of all blocks (n blocks) and calculate the amount of code data of one frame image.

The counter A calculates the code data amount NA when the compression is performed using the first quantization table.
The counter B calculates the code data amount NB when the compression is performed using the second quantization table.

Here, an example is shown in which the scale factor SFa for the first quantization table is larger than the scale factor SFb for the second quantization table. That is, if the first quantization table is used, the second
Higher compression than when using the quantization table of

Code data amount NA when the same image data is compressed using the first quantization table
Is smaller than the code data amount NB when compression is performed using the second quantization table.

In FIG. 1, the scale factor determining unit 7
Is based on two types of code data amounts NA and NB.
And a second candidate scale factor SFx is determined.
The scale factor SFx of the first candidate is the target data amount N
It is estimated as a data compression degree for generating the code data of X.

The scale factor determining section 7 determines the second candidate scale factor SFxx in addition to the first candidate scale factor SFx. In the subsequent processing, when data compression is performed using the first candidate scale factor SFx,
If code data close to the target data amount NX and smaller than the target data amount NX can be generated, the code data using the second candidate scale factor SFxx is not used. However, when data compression is performed using the first candidate scale factor SFx and code data larger than the target data amount NX is generated, it is not possible to use the code data based on the first candidate scale factor SFx. Since it is not appropriate, the second candidate scale factor SFxx is used. Details will be described later.

When the scale factors SFx and SFxx of the first and second candidates are obtained, the statistical processing (step S1 in FIG. 2) ends.

FIG. 5 is a graph for explaining the processing of the scale factor determining section 7. The vertical axis is the scale factor, and the horizontal axis is the code data amount. The scale factor determination unit 7 holds the scale factor as a function of the amount of code data. As an example, a case where the scale factor is proportional to the code data amount will be described.

When compression is performed using the first scale factor SFa, a code data amount NA is obtained. Second
When the compression is performed using the scale factor SFb, the code data amount NB is obtained. Based on these data, characteristics (approximate by a straight line) as shown in FIG. 5 are set.

The target data amount NX is designated externally. A scale factor corresponding to the target data amount is obtained from the above-described characteristics. It can be seen that in order to obtain the target data amount NX, compression should be performed using the first candidate scale factor SFx.

The first candidate scale factor SFx can be obtained from FIG. 5 as follows.

(NB-NA): (NX-NA) = (SFa-SFb): (SFa-SFx) (NB-NA) × (SFa-SFx) = (NX-NA) × (SFa-SFb) ( SFa−SFx) = (NX−NA) × (SFa−SFb) / (NB−NA) SFx = SFa − {(NX−NA) × (SFa−SFb) / (NB−NA)}

Note that, using the above equation, the scale factor S
In addition to obtaining Fx, the scale factor SFx may be obtained by the above approximate expression. In addition, in addition to the calculation formula,
Scale factor SFx using lookup table
May be determined. Further, the relationship between the scale factor and the code data amount is not limited to the proportional relationship, but may be set to another relationship such as an inverse proportionality. However, it is desirable that the scale factor SFx can be determined with a short and simple configuration.

Further, when code data is generated by performing data compression processing using the scale factor SFx, a safety factor is further added to the scale factor obtained above in order to prevent the code data amount from exceeding the target data amount. You may make it hang.

Next, the second candidate scale factor SFx
A method for determining x will be described. The second candidate scale factor SFxx is used when a data amount larger than the target data amount NX is generated with the first candidate scale factor SFx. Second candidate scale factor S
It is necessary to set the second candidate scale factor SFxx so that the code data amount using Fxx is smaller than the code data amount using the first candidate scale factor SFx. That is, when the scale factor is set in a range of 1 or less, the second candidate scale factor SFx
x needs to be set smaller (to a higher data compression degree) than the first candidate scale factor SFx.

The scale factor can be set to a value of 1 or more, but for convenience of explanation, an example in which the scale factor is set to a value of 1 or less will be described below.

The second candidate scale factor SFxx is
Based on the first candidate scale factor SFx obtained above, for example, it can be obtained by the following equation.

SFxx = 0.9 × SFx Further, the second candidate scale factor SF is obtained by another method.
xx can also be determined. For example, the target data amount NX
A target data amount NXX smaller than NXX = 0.9 ×
NX, and using a characteristic line as shown in FIG.
A scale factor corresponding to the target data amount NXX may be obtained and used as the second candidate scale factor SFxx.

Returning to FIG. 1, after the statistical processing is completed, data compression processing (step S2 in FIG. 2) is performed. In the data compression process, the quantization unit 4 inputs the first candidate scale factor SFx as the first scale factor SFa and the second candidate scale factor SFxx as the second scale factor SFb. If compression is performed using the first candidate scale factor SFx, it can be estimated that code data close to the target data amount NX is generated.

The scale factor S of the first and second candidates
Formal data compression processing is performed using Fx and SFxx.
The data compression processing is basically the same as the above-described statistical processing except that the value of the scale factor is different.

The encoding section 5 alternately outputs the code data D based on the first candidate scale factor SFx and the code data D based on the second candidate scale factor SFxx in a time-division manner. Details will be described later with reference to FIGS.

The code data D is stored in the memory 81 or 82 in a time-division manner via the multiplexer 83. The memory 81 has a first candidate scale factor SF.
The first code data according to x is stored. The memory 82 stores the second code data based on the second candidate scale factor SFxx.

The code data amount counting section 6 outputs the code data amounts NA and NB. The code data amount NA is the code data amount when the first candidate scale factor SFx is used. The code data amount NB is a code data amount when the second candidate scale factor SFxx is used.

According to the code data amount NA and the size of NB,
It can be divided into the following cases. In each case, the multiplexer 84 outputs either the first code data stored in the memory 81 or the second code data stored in the memory 82 to the outside as the code data DD. The code data DD is stored in, for example, a memory card.

(1) Code data amount NA is equal to target data amount N
When the compression factor is smaller than or equal to X When the compression is performed using the first candidate scale factor SFx, the code data amount becomes NA. In this case, since the code data amount NA is smaller than or equal to the target data amount NX, the code data based on the first candidate scale factor SFx is employed. That is, among the code data output from the encoding unit 5, the first candidate scale factor SF
The first code data generated by x is output as formal code data DD.

(2) The code data amount NA is equal to the target data amount N
X and the code data amount NB is equal to the target data amount N
When the code data amount is smaller than or equal to X Since the code data amount NA is larger than the target data amount NX, the code data with the first candidate scale factor SFx cannot be adopted. On the other hand, the code data amount NB
Is smaller than or equal to the target data amount NX, the code data based on the second candidate scale factor SFxx is adopted. That is, of the code data output from the encoding unit 5, the second candidate scale factor SFx
The second code data generated by x is output as formal code data DD.

(3) When both the code data amounts NA and NB are larger than the target data amount NX The code data amounts NA and NB are both the target data amount NX
Therefore, neither of the code data based on the first and second candidate scale factors SFx and SFxx can be adopted. In this case, a third candidate scale factor SFx and a fourth candidate scale factor SFx are newly added.
Determine x. Specifically, the scale factor determining unit 7
Determines the scale factors SFx and SFxx of the third and fourth candidates based on the code data amounts NA and NB. Third and fourth candidate scale factors SFx, SFxx
Needs to be smaller than the scale factors of the first and second candidates.

Thereafter, the scale factors SFx and SF of the third and fourth candidates are input to the quantization unit 4 as first and second scale factors SFa and SFb, respectively, and compressed again to generate code data. . Subsequent processing is the same as the above-described case of the first and second candidate scale factors. Still, when code data exceeding the target data amount NX is generated, the above processing is repeated.

Next, the processing procedure of the statistical processing and the data compression processing in the image data compression system of FIG. 1 will be described. The statistical processing and the data compression processing are basically the same processing.

First, the statistical processing will be described. (1) Statistical Processing In statistical processing, the amounts NA and NB of two types of code data are calculated in approximately one data compression process. This is made possible by taking advantage of the overwhelmingly long processing time in the DCT unit 2 in the data compression processing.

The processing time in the DCT unit 2 is 5
It accounts for ~ 80%. The time required for DCT processing (DCT unit 2) and the time required for quantization processing (quantization unit 4) of one block (8 × 8 pixels) of image data will be compared.

The DCT unit 2 performs the following calculations. Here, a case where a normal DCT algorithm is used is shown.

Multiplication 1024 times Addition 896 times The quantization unit 4 performs the following number of calculations.

Multiplication 64 (= 8 × 8) As described above, the number of operations performed by the DCT unit 2 is orders of magnitude larger than that of the quantization unit 4, and the processing time is overwhelmingly long.

FIG. 6 is a timing chart showing the timing at which the image data compression system of FIG. 1 performs statistical processing. The horizontal axis indicates time.

The DCT unit 2 performs processing in the order of processing α1, processing α2, and processing α3. The processing α1 is performed in block 1 (1
This is a DCT process performed on the image data of the (th block). Process α2 is a DCT process performed on image data of block 2 (second block), and process α3 is performed on image data of block 3 (third block).

At time t0, DCT unit 2 starts DCT processing α1 on the image data of block 1. When detecting the end of the DCT processing α1 at time t10, the controller 8 sends the block 2 to the DCT unit 2.
To start the DCT process α2 for the image data. Subsequently, when the end of the DCT process α2 is detected at time t20, the DCT unit 2 is instructed to start the DCT process α3 for the image data of the block 3. Time t
At 30, the DCT process α3 ends. DCT unit 2
Processes image data continuously from block 1 to block n (final block).

Next, the processing of the quantization unit 4 and the encoding unit 5 will be described. At time t10, the controller 8 performs the DCT processing α on the image data I of the block 1.
When the end of 1 is detected, the quantization unit 4 is instructed to start the quantization process β1A for the DCT coefficient F generated by the DCT process α1. The quantization unit 4 performs a quantization process β1A using the first quantization table (scale factor SFa). The quantization process β1A is a process of quantizing the DCT coefficient F of the block 1 using the first quantization table.

At time t11, the controller 8
When the end of the quantization processing β1A is detected, the quantization processing β1
The encoding unit 5 is instructed to start the encoding process γ1A for the quantized data R generated by A. The encoding process γ1A is an encoding process for data of the block 1 using the first quantization table.

At time t12, when the encoding process γ1A is completed, encoded data using the first quantization table is generated. Thereafter, although not shown in FIG. 6, the counter A of the code data amount counting unit 6 (FIG. 1) counts the amount of the code data.

In the above, the quantization process and the encoding process were performed using the first quantization table (scale factor SFa). Next, quantization processing and encoding processing are performed using the second quantization table (scale factor SFb).

At time t12, the controller 8
When the end of the encoding process γ1A is detected, the DCT process α1
Quantization process β1 for the DCT coefficient F generated by
Instruct the quantization unit 4 to start B. However, at this time, the controller 8 instructs to perform the quantization process β1B using the second quantization table. The quantization process β1B is a process for quantizing the DCT coefficient F of the block 1 using the second quantization table.

At time t13, the controller 8
When the end of the quantization process β1B is detected, the quantization process β1
The encoding unit 5 is instructed to start the encoding process γ1B for the quantized data R generated by B. The encoding process γ1B is an encoding process for data of the block 1 using the second quantization table.

At time t14, when the encoding process γ1B is completed, encoded data using the second quantization table is generated. Thereafter, although not shown in FIG. 6, the counter B of the code data amount counting unit 6 (FIG. 1) counts the amount of the code data.

At time t12, the generation of code data by the first quantization table ends, and at time t14, generation of code data by the second quantization table ends. Thereafter, at time t20, the DCT processing α2
Ends. That is, while performing the DCT processing α2, the quantization processing β1A and β1B and the encoding processing γ1A
And γ1B end.

In the above, the quantization process and the encoding process have been performed on the block 1. Next, quantization processing and encoding processing are performed on the block 2.

At time t20, the controller 8
When the end of the DCT process α2 for the image data I of the block 2 is detected, the DT generated by the DCT process α2 is detected.
The quantization unit 4 is instructed to start the quantization process β2A for the CT coefficient F. At this time, the controller 8 instructs to perform the quantization process β2A using the first quantization table. The quantization process β2A is performed by the DC
This is a process for quantizing the T coefficient F using the first quantization table.

At time t21, the controller 8
When the end of the quantization process β2A is detected, the quantization process β2
The encoding unit 5 is instructed to start the encoding process γ2A for the quantized data R generated by A. The encoding process γ2A is an encoding process for data using the first quantization table for block 2.

At time t22, when the encoding process γ2A ends, encoded data using the first quantization table is generated. Thereafter, although not shown in FIG. 6, the counter A of the code data amount counting unit 6 (FIG. 1) counts the amount of the code data.

Next, quantization processing and encoding processing are performed using the second quantization table. The controller 8 sets the time t
At 22, when the end of the encoding process γ2A is detected,
The quantization unit 4 is instructed to start the quantization process β2B for the DCT coefficient F generated by the DCT process α2. However, at this time, the controller 8 instructs to perform the quantization process β2B using the second quantization table. The quantization process β2B is a process for quantizing the DCT coefficient F of the block 2 using the second quantization table.

At time t23, the controller 8
When the end of the quantization process β2B is detected, the quantization process β2B
The encoding unit 5 is instructed to start the encoding process γ2B for the quantized data R generated by B. The encoding process γ2B is an encoding process for data on the block 2 using the second quantization table.

At time t24, when the encoding process γ2B is completed, encoded data using the second quantization table is generated. Thereafter, although not shown in FIG. 6, the counter B of the code data amount counting unit 6 (FIG. 1) counts the amount of the code data.

At time t22, the generation of code data using the first quantization table ends, and at time t24, generation of code data using the second quantization table ends. Thereafter, at time t30, the DCT processing α3
Ends. While the DCT process α3 is being performed, the quantization processes β2A and β2B and the encoding processes γ2A and γ2B are completed.

Hereinafter, similarly, the processing is continued up to the last block n. The time for this series of statistical processing corresponds to the time for the DCT unit 2 to perform the DCT processing on all the blocks, that is, substantially the time for one ordinary data compression process.

Although the time for the statistical processing is the time for one data compression processing, the image data compression system uses the first data compression processing.
And the second quantization table, two types of code data are generated, and the code data amounts NA and NB can be calculated.

FIG. 6 shows the timing in each processing unit. Next, the processing timing for each block will be described.

FIG. 7 is a timing chart showing the timing of performing the statistical processing in units of blocks. The horizontal axis indicates time and corresponds to the time in FIG.

First, the processing of block 1 will be described. At time t0, DCT processing α1 for the image data of block 1 starts, and at time t10,
The CT processing α1 ends. When the DCT process α1 ends, a code data generation process 51A using the first quantization table is performed, and then a code data generation process 51B using the second quantization table is performed.

The process 51A includes a quantization process β1A using the first quantization table and a coding process γ1A of the quantized data generated by the quantization process β1A. The process 51B includes a quantization process β1B using the second quantization table and an encoding process γ1B of the quantized data generated by the quantization process β1B.

Next, the processing of block 2 will be described. At time t10, DCT processing α2 for the image data of block 2 starts, and at time t20,
The DCT process α2 ends. When the DCT process α2 ends, a code data generation process 52A using the first quantization table is performed, and then a code data generation process 52B using the second quantization table is performed. Process 51A is
It includes a quantization process β2A and an encoding process γ2A. Process 51
B includes a quantization process β2B and an encoding process γ2B.

As described above, since the DCT processing time is long, two quantization processing and coding processing (for example, 51A and 51B) can be performed during DCT processing (for example, α2) for one block. it can. Thereby, substantially two times of statistical processing can be performed during almost one data compression processing for one frame image. One statistical process uses the first quantization table, and calculates the code data amount NA. Another statistical process is
The second quantization table is used, and the code data amount NB is calculated.

In the statistical processing, after calculating the code data amounts NA and NB, the scale factors SF of the first and second candidates are calculated.
x and SFxx are obtained. Next, the data compression processing will be described. (2) Data compression processing When the statistical processing is completed, data compression processing is subsequently performed using the first and second candidate scale factors SFx and SFxx. The data compression processing is the same as the above-described statistical processing except that the value of the scale factor is different.
As with the statistical processing, compression of one frame of image data is performed. By the data compression, the first candidate scale factor S
Code data using Fx and code data using the second candidate scale factor SFxx are generated.

If the scale factor of the first candidate is an appropriate value, code data using the scale factor of the first candidate is adopted. When the scale factor of the first candidate is not appropriate, it is determined whether the scale factor of the second candidate is appropriate. If the second candidate scale factor is appropriate, code data using the second candidate scale factor is adopted. If the scale factors of the first and second candidates are both inappropriate, a new scale factor is determined and the data compression process is performed again.

FIG. 8 shows the code data amount counting section 6 in FIG.
Here is another example. The code data amount counting unit 6 counts the code data amount NA using the first quantization table (scale factor SFa) and the code data amount NB using the second quantization table (scale factor SFb).

The code data counting section 6 includes a counter 61
And a register A and a register B. The counter 61
Under the control of the controller 8 of FIG. 1, the amount of code data using the first quantization table and the amount of code data using the second quantization table are counted in a time division manner.

The amount of code data in block units using the first quantization table is added to the register A.
When the addition for all the blocks is completed, the register A stores the code data amount NA. The code data amount NA is the code data amount of one frame image when the first quantization table is used.

On the other hand, to the register B, the code data amount of the second quantization table in block units is added. When the addition for all the blocks is completed, the register B stores the code data amount NB. The code data amount NB is the code data amount of one frame image when the second quantization table is used.

FIG. 9 shows another example of the quantization table 4 of FIG. The quantization unit 4 includes a quantization table memory 11,
13 and 14, a multiplier 12, and a divider 15.

The memory 11 stores a reference quantization table Q (FIG. 3). The multiplier 12 has a reference quantization table Q
Is multiplied by a scale factor SF. That is, the scale factor SF is assigned to all elements of the matrix of the reference quantization table Q.
Multiply by The multiplier 12 outputs a quantization table SF × Q. The quantization table SF × Q is stored in either the memory 13 or the memory 14.

The memory 13 stores a first quantization table, and the memory 14 stores a second quantization table. The scale factor SF is different between the first quantization table and the second quantization table. The data compression degree of the code data is determined by the value of the scale factor SF. The first quantization table is represented by Q × SFa, and the second quantization table is represented by Q × SFb.

Either the first quantization table or the second quantization table is supplied to the divider 15 as SF × Q.

As shown by the following equation, the divider 15
The DCT coefficient Fuv stored in the CT coefficient memory 3 (FIG. 1) is divided by a quantization table SF × Quv to output a quantization coefficient Ruv. Rounding round means rounding to the nearest integer.

Ruv = round [Fuv / (SF × Quv)] Unlike the quantization unit 4 of FIG. 9, the quantization unit 4 of FIG. 1 stores a value obtained by multiplying the quantization table Q by the scale factor SF. Does not have the two memories 13 and 14. The quantizing unit in FIG. 1 can reduce the memory capacity as compared with the quantizing unit in FIG. 9, so that the size and cost of the system can be reduced.

However, in the quantizing section of FIG. 1, the multiplexer 71 supplies the scale factor SF to the multiplier 12 and
After performing the operation in the multiplier 12, Q × SFa or Q × SFa
SFb is supplied to the divider 15. That is, the multiplier 12
The timing is delayed by one cushion. It is necessary to adjust the timing delay.

On the other hand, in the quantization unit of FIG. 9, the data read from the quantization table memory 13 or the quantization table memory 14 is directly supplied to the divider 15,
Adjustment of timing is easy, and there is no delay in calculation.

Next, an example in which statistical processing is performed only on sample blocks will be described. In the above-described statistical processing, processing was performed on all blocks of the image of one frame. However, since the statistical processing is only for estimating the amount of code data, it is not always necessary to perform the processing for all blocks. Therefore, processing is not performed for all blocks, but is performed only for sample blocks, so that processing time can be reduced.

FIGS. 10A to 10C show sample examples of sample blocks. The figure shows a collection of two-dimensional blocks. The hatched blocks are sample blocks.

FIG. 10A shows a checkerboard-shaped sample block. FIG. 10B shows a sample block in the form of a vertical stripe. FIG. 10C shows a horizontal stripe-shaped sample block. For example, in the case of FIG. 10A, statistical processing is performed on every other block in the order of block 1, block 3, block 5,.

[0127] These sample blocks are half the number of all blocks. If the statistical processing is performed only on these sample blocks, the time required for the statistical processing can be reduced to about half. However, the amount of code data calculated by the statistical processing is half of the amount of code data of one frame image. Considering that the code data amount is half,
It is necessary to determine scale factors SFx and SFxx.

In the statistical processing, a case has been described in which two types of quantization tables are used. However, three or more types of quantization tables may be used. When processing is performed using three or more scale factors, the memories 81 and 8 in FIG.
2 and the number of memories 13 and 14 in FIG. 9 are also three or more.

However, it is desirable that the quantization processing and the encoding processing be completed within the DCT processing time for one block. Normally, if two types of quantization tables are used, they will fit within the DCT processing time.
If the type of quantization table used is increased, statistical processing with high precision, that is, a more optimal scale factor SF
x, SFxx can be obtained.

Also, as a method of adjusting the data compression degree, a method of changing the scale factor has been described. However, a method of changing the quantization table itself regardless of the scale factor may be used, or another method may be used.

According to the plurality of embodiments, statistical processing can be performed substantially two or more times within the time of one data compression processing for substantially one frame image.
In the statistical processing, an optimum scale factor can be obtained at high speed and with high accuracy. Image data compression system
Code data of a target data amount can be generated quickly and accurately.

Further, the first and second data are statistically processed.
Candidate scale factors can be determined. If the scale factor of the first candidate is appropriate, code data based on the scale factor of the first candidate may be adopted. First
Even when the scale factor of the candidate is inappropriate, it is possible to generate appropriate code data using the scale factor of the second candidate.

Although the present invention has been described in connection with the preferred embodiments,
The present invention is not limited to these. For example, it will be apparent to those skilled in the art that various modifications, improvements, combinations, and the like can be made.

[0134]

As described above, according to the present invention,
After the DCT means has finished generating the DCT coefficients of the first block, the code data generating means generates the code data at two types of data compression, and the DCT means generates the DCT coefficients of the second block. The steps can be performed in parallel. By parallelizing the process, the speed of the data compression process can be increased. Also, by generating code data with at least two types of data compression,
High-precision fixed length processing can be performed.

Further, by estimating the third data compression degree for generating the target code data amount and the fourth data compression degree higher than the third data compression degree, the third or the third data compression degree is estimated. Formal compression can be performed using any appropriate one of the four data compression degrees. When the third data compression degree generates more code data than the target code data amount, it is possible to employ a smaller code data using the fourth data compression degree.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of an image data compression system according to an embodiment of the present invention.

FIG. 2 is a flowchart illustrating a processing procedure performed by the image data compression system according to the embodiment.

FIG. 3 is a diagram illustrating an example of a reference quantization table of FIG. 1;

FIG. 4 is a diagram illustrating a method in which the code data amount counting unit of FIG. 1 counts the code data amount.

FIG. 5 is a graph for explaining processing of a scale factor determination unit in FIG. 1;

FIG. 6 is a timing chart showing the timing of each processing unit in which the image data compression system performs processing.

FIG. 7 is a timing chart showing the timing of each block in which the image data compression system performs processing.

FIG. 8 is a diagram illustrating another example of the code data amount counting unit in FIG. 1;

FIG. 9 is a diagram illustrating another example of the quantization unit in FIG. 1;

FIG. 10 shows a sample example of a sample block. FIG.
0 (A) is a checkered sample block, FIG.
(B) is a vertical striped sample block, FIG.
(C) is a diagram showing a sample block in the form of a horizontal stripe.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Image memory 2 Discrete cosine transform (DCT) part 3 DCT coefficient memory 4 Quantization part 5 Encoding part 6 Code data amount counting part 7 Scale factor determination part 8 Controller 11 Reference quantization table memory 12 Multiplier 13 First quantum Quantization table memory 14 second quantization table memory 15 divider 61 counter 71 multiplexer 81 first code data memory 82 second code data memory 83,84 multiplexer

Claims (7)

[Claims] An image data supply means (1) for supplying image data by dividing the image data into blocks, and a discrete cosine transform (DCT) of the block-wise image data supplied from the image data supply means. DCT means (2) for generating a DCT coefficient for each block; and code data for generating code data at the first and second data compression degrees for the DCT coefficient each time the DCT means generates a DCT coefficient for one block. Generating means (4, 5); and a counter (6) for accumulating the amount of code data for each of the first and second data compression degrees generated by the code data generating means for all supplied blocks. And code data of a target data amount according to both the code data amounts of the first and second data compression degrees generated by the counter. Image data having a third data compression degree considered to be suitable for generation and a data compression degree estimating means (7) for estimating a fourth data compression degree higher than the third data compression degree Compression system. 2. The apparatus according to claim 1, further comprising means for instructing said code data generation means to generate two types of code data with the third and fourth data compression degrees estimated by said data compression degree estimation means. Image data compression system as described. 3. The code data generating means includes a quantizing means for quantizing DCT coefficients, and includes first and second code data.
3. The image data compression system according to claim 1, wherein the code data of the data compression degree is generated by performing quantization using the first and second quantization tables, respectively. 4. The image according to claim 3, wherein said code data generating means includes means for generating first and second quantization tables by multiplying a reference quantization table by first and second scale factors. Data compression system. 5. The image data compression system according to claim 3, wherein said code data generation means includes coding means for performing Huffman coding after quantizing DCT coefficients. 6. The DCT means for performing DCT on image data in units of blocks, wherein the code data generating means performs one of the following steps before the DCT means completes DCT of image data for a certain block.
The image data compression system according to any one of claims 1 to 5, wherein the generation of the code data of the first and second data compression degrees for the immediately preceding block is completed. 7. The image data compression system according to claim 1, wherein said image data supply means supplies image data of a sample block in an image of one frame.