JPH08116452A - Data compressor - Google Patents

Abstract

PURPOSE: To attain high speed data compression while keeping a high compression rate by collecting results of learning by plural data compression means and allowing them to be shared in common. CONSTITUTION: When image information being compression object data is fed to an image split means 101., the image information is divided into plural areas and they are processed in parallel by arithmetic coding processing means 102a-l02n. The arithmetic coding processing means 102a-102n use a learning algorithm to vary a compression parameter and the result of learning by the arithmetic coding processing means 102a-102n are collected through a symbol occurrence probability calculation means 105 and a symbol occurrence probability setting means 107 and the compression parameter based on the result of collection is fed back to the arithmetic coding processing means.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a data compression device suitable for storage and transmission of digital image signals.

[0002]

2. Description of the Related Art Various techniques have been proposed in order to speed up encoding / decoding processing of image information and the like. For example,
In Japanese Unexamined Patent Publication No. 63-74169, a plurality of encoding / decoding processing units are provided, one image information is divided into a plurality of pieces, and the divided image information is provided to separate encoding / decoding processing units. Dispensing techniques are disclosed. That is, by performing encoding / decoding processing on each divided image information in parallel, the amount of image information that can be encoded or decoded per unit time is increased as a whole, and the encoding / decoding is performed. It is to speed up the process.

Further, there are known techniques for using various learning algorithms in order to efficiently encode image information. Here, the learning algorithm is to analyze the property of the data to be compressed (the pattern of the appearing symbols, etc.) while compressing the data to be compressed based on the compression parameter in the initial state, and then to obtain a high compression rate. So that the compression parameter is set based on the analysis result. When the learning algorithm is used, the learning content does not converge at the initial stage of the compression process, and a high compression rate cannot be obtained. For example, it is assumed that the image information shown in FIG. 13A is the compression target data and that the hatched portion is compressed until the compression parameter obtained by the learning algorithm converges to a value suitable for the compression target data. Then, the compression rate of this portion becomes low.

[0004]

By the way, Japanese Patent Laid-Open No. Sho 63-63
If the learning algorithm is simply applied to the technique described in Japanese Patent Laid-Open No. 74169, there arises a problem that the compression rate is extremely deteriorated. For example, assuming that the image information is divided into four areas as shown in FIG. / Since the decoder performs learning independently, the area of the part where the compression rate becomes low (hatched part) is
It is about 4 times that of (a). The present invention has been made in view of the above-mentioned circumstances, and an object of the present invention is to provide a data compression device having a high speed and a high compression rate.

[0005]

In order to solve the above-mentioned problems, in the structure according to claim 1, the input data is compressed while the compression parameter is appropriately changed based on a predetermined learning algorithm. A plurality of data compressing means for outputting and a data dividing means for dividing the data to be compressed when the data to be compressed are inputted, and a totaling for summing learning results of the data compressing means. Means and a compression parameter setting means for setting a common compression parameter for each of the data compression means based on the result of the totalization by the totalizing means.

According to the second aspect of the invention, the first and second data compression means compresses and outputs the input data while appropriately changing the compression parameter based on the learning algorithm. When the data to be compressed is input, the data to be compressed is divided and supplied to the first and second data compressing means, and the first to second data compression means.
Of the compression parameter of the second data compression unit based on the learning result of the activation control unit that activates the second data compression unit and then activates the second data compression unit. And compression parameter setting means for setting an initial value.

According to a third aspect of the present invention, in the data compression apparatus according to the first aspect, the data to be compressed is composed of a plurality of types of symbols, and each of the data compression means is , A symbol having a high appearance probability is converted into binary data having a short bit length, and a symbol having a low appearance probability is converted into binary data having a long bit length, thereby compressing the supplied data. Appearance probability calculating means for sequentially calculating the appearance probability of the symbols included in the compression target data, and it is determined whether or not the variation amount of the appearance probability satisfies a predetermined convergence condition, and if the convergence condition is satisfied, the compression is performed. Convergence condition determination means for stopping the operation of the parameter setting means is provided.

[0008]

In the structure according to the first aspect, the plurality of data compression means compresses the input data and outputs it while appropriately changing the compression parameter based on a predetermined learning algorithm. Further, when the data to be compressed is input, the data dividing means divides the data to be compressed and supplies it to each of the data compressing means. At that time, the totaling means totals the learning results of each data compressing means, and the compression parameter setting means sets a common compression parameter for each data compressing means based on this totaling result. As a result, the learning result in each data compression means is shared.

According to the second aspect of the invention, the data dividing means divides the data to be compressed into first and second data.
Data compression means. When the activation control means activates the first data compression means, the first data compression means activates the first data compression means.
The data compressing means compresses the input data and outputs it while appropriately changing the compression parameter based on a predetermined learning algorithm. Next, the compression parameter setting means sets the initial value of the compression parameter in the second data compression means, based on the learning result in the first data compression means. Next, when the second data compression means is activated by the activation control means, the second data compression means compresses the input data and outputs it while appropriately changing the compression parameter based on a predetermined learning algorithm. To do.
In this way, the learning result of the first data compression means is used in the second data compression means.

Further, in the structure according to the third aspect, the appearance probability calculating means successively calculates the appearance probability of the symbols included in the compression target data. The convergence condition determination means determines whether or not the variation amount of the appearance probability satisfies a predetermined convergence condition,
When this convergence condition is satisfied, the operation of the compression parameter setting means is stopped. As a result, each data compression unit independently learns when the variation in the occurrence probability converges to some extent.

[0011]

【Example】

A. Outline of each embodiment First, an outline of each embodiment will be described. First, in the first embodiment, the image information to be compressed is divided into a plurality of blocks as shown in FIG. Each block is separately encoded by n arithmetic encoding processing means. When the coding of these blocks is completed, the frequency of appearance of each symbol in each block is aggregated, and the result is set as a learning result for each arithmetic coding processing means.
Therefore, the amount of data until the learning result converges can be suppressed to the same extent as when one arithmetic coding processing unit is used. Moreover, since the actual encoding process is performed in parallel by the n arithmetic encoding processing means, it is extremely fast.

In the second embodiment, the image information to be compressed is divided into a plurality of blocks as shown in FIG. First, Block 1 is executed by one arithmetic coding processing means.
-1 is coded, and the frequency of appearance of each symbol in the same block is obtained in the process. Next, using the result, Block 1- is performed by the two arithmetic coding processing means.
2, Block 2-1 are coded in parallel, and the frequency of appearance of each symbol is aggregated. Next, using the result, Block 1-3 and Block2- are provided by the three arithmetic coding processing means.
2, Block 3-1 is encoded in parallel. And so on
Other blocks are also processed. As a result, the learning result in one arithmetic coding processing means is reflected in the other arithmetic coding processing means, and the coding processing can be performed at high speed and high compression rate.

In the third embodiment, the image information to be compressed is divided into a plurality of blocks as shown in FIG. First, a block is created by one arithmetic coding processing unit.
A is encoded, and the frequency of appearance of each symbol is obtained in the process. Next, using the result, the areas [1] to [N] are coded in parallel by the n arithmetic coding processing means. As a result, the learning result in Block A is also used in other arithmetic coding processing means, and the coding processing can be performed at high speed and with a high compression rate.

B. First embodiment B-1. Arithmetic Coding Method Hereinafter, an arithmetic coding method adopted in this embodiment (and other embodiments described later) will be described. The arithmetic coding method will be described in detail below, but in a nutshell, the probability line is divided into sections according to the appearance of the input symbol sequence, and a binary decimal value indicating the position within the divided section. Is the code for the sequence, and it has this name because it is constructed sequentially by arithmetic operations after the code. This arithmetic coding method is JBI, which is a joint standard proposal between the International Telegraph and Telephone Advisory Committee and the International Organization for Standardization.
International standardization work is underway in G (Joint Bi-level Image experts Group), which is a basic method of QM-Coder adopted as a binary image coding method.

Hereinafter, the appearance probabilities of the two symbols "0" and "1" are P ₀ = 0.75 and P ₁ = 1-P ₀ =, respectively.
An example of encoding processing by the binary arithmetic encoding method when given by 0.25 will be described with reference to FIG. In this example, the symbol “0” is a symbol having a high appearance probability and is called a dominant symbol, and the symbol “0” is a symbol having a low appearance probability and is called a inferior symbol. That is, the dominant symbol that appears frequently is given a wide area on the number line. In FIG. 11, first, the number line of 0 or more and less than 1 is divided into 0.75: 0.25, and 0 or more and 0.7
An area A _{0 of} less than 5 and an area A ₁ of 0.75 or more and less than ₁ are created. Next, if the first symbol input from the information source is "0", the area A0 is selected, and if it is "1", the area A1 is selected.

For example, when the first symbol input from the information source is "0" and the area A0 is selected, the selected area A0 is similarly subdivided into 0.75: 0.25, a ₀₀ to create a (0 or 0.5625 (0.75 ²⁾ below) and a region a ₀₁ (less than 0.5625 or 1). Here, if the next input symbol is “0”, the area A
_{If 00} is "1", area A ₀₁ is selected. Such area division / area selection processing is performed for all input symbols, and the binary decimal representation of one point in the area obtained by the last division is used as a code. As the one point in this area, the lowest point in the selected area is usually used.

In FIG. 11, the area surrounded by the thick line is
The encoding process when the input symbol sequence "0010" is encoded is shown, and the value of one point shown as "Code" in the figure is used as a code generated as a result of encoding this input symbol sequence. To be Table 1 shows the result of coding processing of an input symbol sequence consisting of 4 bits by this binary arithmetic coding method. In addition, B at the end of the numerical values in Table 1
Indicates that it is a binary number. In Table 1, the "probability of appearance" and the "bottom point in the area" are shown for each input symbol series. Here, the significant digits of the "bottom point in the area" corresponding to each input symbol series are rounded down, leaving the minimum necessary to distinguish them from the bottom points of other areas. The bit string of the part is "sign"
And

[0018]

[Table 1]

According to Table 1, the input symbol sequence "000"
It can be seen that the code data length generated when 0 "," 0001 ", and" 0010 "are coded is smaller than the input symbol sequence length, and the coding process is effective. The code data length generated when the input symbol sequence other than is coded is equal to or longer than the input symbol sequence length. It can be mentioned that the values of the appearance probabilities of the two symbols “0” and “1” used in the encoding process are not optimal for these input symbol sequences. The reason is that the length of the input symbol sequence to be processed was not sufficiently long, that is, in the above example, the bit length of the input symbol sequence was set to 4 bits in order to simplify the description, but this length is about this. Is enough It is difficult to obtain the effect.

That is, when the values of the appearance probabilities of the optimum symbols "0" and "1" are used for the input symbol sequence, and the input symbol sequence has a sufficient length, this arithmetic coding method is used. It is known that the effect of data reduction becomes larger, and in such a case, it is theoretically possible to compress to a code length close to the entropy value of the input data sequence.

Further, in the above example, the case where the appearance probability value of the input symbol is constant has been described. However, the arithmetic coding method is adapted to the local property of the input symbol sequence while performing the coding process. It has a feature that the appearance probability value of the input symbol can be easily changed dynamically. In the arithmetic coding method, if the same appearance probability value of the input symbol is used during the encoding process and the decoding process, there is no problem even if the appearance probability value of the input symbol is changed every time one symbol is encoded. . Therefore, the appearance probability value of each symbol is obtained from the symbol sequence that has already been encoded, and the appearance probability value of the input symbol is sequentially changed based on it, so that the encoding process is performed using the optimum symbol appearance probability value. It becomes possible to use the learning algorithm of doing. In the QM-Coder described above,
Prepare the appearance probability values of multiple symbols in advance,
The configuration is such that this is switched intermittently during the encoding process.

When image information is coded by the above-mentioned arithmetic coding method, input pixel data is sequentially input as a symbol series, the appearance probability of each symbol is predicted based on the state of surrounding pixel data, and the prediction result thereof is obtained. Based on the above, the above-mentioned encoding processing is performed. At the start of the encoding process, a constant initial value is set as the probability of appearance of each symbol. Since the appearance probability of each symbol of the image information to be processed is unknown at the start of the encoding process, the process is usually started assuming that the appearance probabilities of the respective symbols are equal.

In FIG. 12, in the binary arithmetic coding system,
As the encoding process progresses, it is shown that the initial value of the appearance probability of the inferior symbol set at the start of the encoding process approaches the appearance probability P of the inferior symbol that matches the property of the processing target image information. As shown in FIG. 12, for a certain period of time until the initial value of the appearance probability of each symbol set at the start of the encoding process approaches the appearance probability that matches the property of the input image information to be processed, the input symbol sequence is Since the optimum appearance probability value is not used, the generated code length becomes long, and thus the coding efficiency decreases.

B-2. Configuration of Embodiments Overall Configuration of Embodiments A first embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram of a copying machine according to this embodiment. In the figure, 1 is an automatic document feeder, 2 is an image reading device,
3 is a selector, 4 is an encoding processing unit, 5 is a page memory,
6 is a bus arbiter, 7 is a DMA controller, 8 is an image storage device, 9 is an external input / output I / F unit, 10 is a decoding processing unit, 11 is a multiplexer, 12 is an automatic paper binding device,
13 is an image recording device, 14 is a user interface, 1
5 is a communication controller, 16 is a CPU, 17 is a RAM,
Reference numeral 18 is a PROM, and 19 is an interrupt controller.

The image input means comprises an automatic document feeder 1 and an image reading device 2, and the image reading device 2 is a CCD.
A document is converted into electrical image data and read by an image pickup device such as. The automatic document feeder 1 sequentially sends out documents to a predetermined reading position of the image reading device 2, and discharges them after the reading is completed. The selector 3 is used to switch which of the image data read by the image input means and the image data input from the external input / output I / F unit and stored in the page memory 5 is input to the encoding processing unit 4. .
The encoding processing unit 4 is used to encode the image data to increase the number of images that can be stored in the image storage device 8 described later and to reduce the data transfer speed to the image storage device 8.

In the page memory 5, since the data output speed of the encoding processing unit 4 is not constant, it takes time to start the data transfer of the image storage device 8 and the transfer speed is also not constant. It is used to temporarily store the data output from. The external input / output I / F unit 9 is used to input / output image data via a network line. The image data input / output by the external input / output I / F unit 9 is temporarily stored in the page memory 5. The decoding processing unit 10 decodes the encoded image data. The multiplexer 11 is used to switch whether the output destination of the image data decoded by the decoding processing unit 10 is the page memory 5 or the image output means. The image output means comprises an image recording device 13 and an automatic paper binding device 12,
The image recording device 13 records the decoded image data on a sheet by an electrophotographic recording method using a laser or the like. The automatic document binding device 12 binds sheets according to the number of copies that the image recording device 13 sequentially outputs.

The DMA controller 7 has 6 channels, and the channel 1 from the page memory 5 to the encoding processing unit 4
To the page memory 5, DMA transfer from the encoding processor 4 to the page memory 5, channel 3 DMA transfer from the page memory 5 to the decoding processor 10, channel 4 from the decoding processor 10 to the page memory DMA to 5
Transfer, channel 5 is page memory 5 and image storage device 8
DMA transfer between channels, and the channel 6 performs DMA transfer between the page memory 5 and the external input / output I / F unit 9. The bus arbiter 6 includes an encoding processing unit 4, an image storage device 8, and external input / output I / O.
It arbitrates six types of DMA transfer requests from four devices of the F unit 9 and the decoding processing unit 10 in a predetermined priority order and a predetermined data unit.

The above is concerned with the image data, and in order to operate at a higher speed, it is separated from the CPU bus of the control system described later. The control means is composed of each device connected to the CPU bus of the control system. The CPU 16 is a main control device and controls each device connected to the CPU bus. PR
The OM 18 stores software for operating the CPU 16. The RAM 17 is used as a working memory for the CPU 16. The interrupt controller 19 includes an encoding processing unit 4, an image storage device 8, an external input / output I / F unit 9, a decoding processing unit 10, and a 3-channel communication controller 1.
Accepts the interrupt request from 5 with a predetermined priority and C
PU16 is prompted to process. The communication controller 15 has three channels and is a device for transmitting / receiving commands / status between the CPU and the image input unit, the image output unit, and the user interface 14. The user interface 14 is a device for inputting an operator's instruction and displaying the state of the image processing device.

Configuration of Main Part of Embodiment (Encoding Processor 4) Next, a detailed configuration of the encoding processor 4 will be described with reference to FIG. In the figure, 101 is an image dividing means, which receives the image information to be encoded through the selector 3 and divides this image information into a plurality of blocks as shown in FIG. Thereafter, in FIG. 8, Block 1-1, Block 1-2, ..., Block 1-m are areas [1], Block 2-1, Block 2-2, ..., Block 2-m are areas. [2], ::: Block n-1, Block n-2, ..., Block nm are referred to as a region [N].

Next, 102a to 102n are "N" arithmetic coding processing means, and the image information of the above-mentioned area [1] to area [N] is divided from the image dividing means 101 in units of blocks. Receive and perform encoding processing. Then, each of the arithmetic coding processing means 102a to 102n outputs the coded data obtained as a result of the coding processing and the symbol appearance frequency information indicating the appearance frequency of each symbol in the Block which is the processing target. . Further, each of the arithmetic coding processing means 102a to 102n outputs an image input request signal to the image dividing means 101 when the coding processing can be performed on a new Block.

In response to the image input request signal, the image dividing means 101 sequentially outputs the image information of the corresponding area to the arithmetic coding processing means 102a to 102n. 103
Is an encoding operation control means, which outputs a command to other components as appropriate in order to proceed with the encoding process.
Reference numeral 107 is a symbol appearance probability setting means, which is under the control of the encoding operation control means 103, each arithmetic encoding processing means 10.
Symbol appearance probability value P_Set for 2a to 102n
Set. That is, the symbol appearance probability setting means 10
7 indicates that at the beginning of the encoding process, a predetermined symbol appearance probability initial value P_Set_Init is set to this symbol appearance probability value P.
_Set as arithmetic coding processing means 102a to 102n
Set to. On the other hand, after that, the symbol appearance probability determination result P_Det (details will be described later) calculated based on the aggregation result is set as the symbol appearance probability value P_Set.

Next, reference numeral 104 denotes an encoding processing start control means. When a predetermined command is supplied from the encoding operation control means 103, each of the arithmetic encoding processing means 102a-102.
n is activated to start the encoding process. Reference numeral 105 is a symbol appearance probability calculation means, and each arithmetic coding processing means 10
When the symbol appearance frequency information for each Block is supplied from 2a to 102n, these are aggregated and the aggregated result is output as the symbol appearance probability calculation result P_Tmp. It should be noted that the aggregation result referred to here is the sum of newly supplied symbol appearance frequency information and past supplied symbol appearance frequency information, and calculates the appearance frequency of each symbol within the currently known range. It was done.

By the way, in the present embodiment, as the possible values of the symbol appearance probability value P_Set, several kinds (2 ^-1 , 2 ^-2 , 2 ^-3 , ...) Of the symbol appearance probability values are calculated beforehand. A value stored in a table in 105 and closest to the actual appearance probability of each symbol (symbol appearance probability calculation result P_Tmp) is selected from the table. Then, the set of these selected values is the symbol appearance probability determination result P
_Det is output to the symbol appearance probability setting means 107.

Next, reference numeral 106 is a symbol appearance probability convergence determination means, which stores the history of the symbol appearance probability calculation result P_Tmp, and when a new symbol appearance probability calculation result P_Tmp is supplied, this and the previously supplied symbol. The appearance probability calculation result P_Tmp is compared to determine whether or not the difference between the two meets a predetermined convergence condition. That is, when the difference in the symbol appearance probability calculation result P_Tmp for each symbol is within a predetermined threshold, it is determined that the calculation results have converged. In other words, the converged symbol appearance probability calculation result P_Tmp and the symbol appearance probability determination result P_Det determined based on the converged symbol appearance probability calculation result P_Tet match the characteristics of the input image information to be processed.

When the symbol appearance probability convergence determination means 106 determines that "the calculation result of the appearance probability value of the symbol does not converge to the appearance probability value that matches the property of the input image information to be processed", The calculated symbol appearance probability determination result P_Det is the arithmetic coding processing means 102.
It is set to the symbol appearance probability value P_Set in a to 102n. That is, the encoding operation control means 103 instructs the symbol appearance probability setting means 107 to assign the symbol appearance probability value P_S to each of the arithmetic encoding processing means 102a to 102n.
At the same time as instructing et to be set, the encoding processing start control means 104 is instructed to activate the encoding processing in each of the arithmetic encoding processing means 102a to 102n.

When the symbol appearance probability setting means 107 receives the above instruction, the symbol appearance probability determination result P_Det input from the symbol appearance probability calculating means 105 is set to the symbol appearance probability value P_Set in each of the arithmetic coding processing means 102a to 102n. Set to. Then, the symbol appearance probability value P_Set by the symbol appearance probability setting means 107
When the setting of is completed, the encoding process start control means 104
Outputs an encoding process activation signal to each of the arithmetic encoding processing means 102a to 102n and instructs the arithmetic encoding processing means 102a to 102n to start the encoding process.

On the other hand, the symbol appearance probability convergence determining means 10
6, when it is determined that the symbol appearance probability calculation result P_Tmp has converged to the appearance probability value that matches the property of the input image information, the symbol appearance probability convergence determination means 106 instructs the encoding operation control means 103 to perform the symbol appearance probability. Output a convergence judgment signal. Symbol appearance probability convergence determination means 1
When the symbol appearance probability convergence determination signal is input from 06, the encoding operation control means 103 does not instruct the symbol appearance probability setting means 107 to set the symbol appearance probability value P_Set, and each arithmetic operation Encoding processing means 1
The encoding process start control unit 104 is instructed to start the encoding process in 02a to 102n. As a result, the encoding processing start control means 104 causes the arithmetic encoding processing means 1 to operate.
An encoding process start signal is output to 02a to 102n to instruct the start of the encoding process.

(Decoding Processing Unit 10) Next, the details of the decoding processing unit 10 will be described with reference to FIG. 1 in the figure
Reference numerals 08a to 108n are "N" arithmetic decoding processing means, and each arithmetic code in the encoding processing unit 4 is the encoded data of the area [1] to area [N] described above in units of each Block. It receives from the encoding processing means 102a to 102n, decodes these encoded data into image information, and outputs the image information. Each of the arithmetic decoding processing means 108a to 108n has the image information obtained as a result of the decoding processing and the Block which is the processing target.
And the symbol appearance frequency information indicating the appearance frequency of each symbol in the. Further, 114 is an image synthesizing means, which synthesizes and outputs the image information output from each of the arithmetic decoding processing means 108a to 108n.

Decoding operation control means 109 outputs commands as appropriate to other components in order to proceed with the decoding process. A symbol appearance probability setting unit 113 sets a symbol appearance probability value P_Set for each of the arithmetic decoding processing units 108a to 108n under the control of the decoding operation control unit 109. That is, the symbol appearance probability setting means 113 stores the same symbol appearance probability initial value P_Set_Init as that stored in the above-mentioned symbol appearance probability setting means 107, and at the beginning of the decoding process, this symbol appearance probability is set. Initial probability
P_Set_Init is set in each arithmetic decoding processing means 108a to 108n as this symbol appearance probability value P_Set.
On the other hand, after that, the symbol appearance probability determination result P_Det calculated based on the aggregation result is set as the symbol appearance probability value P_Set.

Next, 110 is a decoding process start control means, and when a predetermined command is supplied from the decoding operation control means 109, each arithmetic decoding processing means 108a-108.
n is activated to start the encoding process. Reference numeral 111 is a symbol appearance probability calculating means, which is configured similarly to the above-mentioned symbol appearance probability calculating means 105. That is, when the symbol appearance probability calculation means 111 is supplied with the symbol appearance frequency information for each Block from each of the arithmetic decoding processing means 108a to 108n, these are aggregated and the symbol appearance probability calculation result P_Tmp and the symbol appearance probability determination result are collected. P_Det
And output. Further, 112 is a symbol appearance probability convergence determination means, which is configured similarly to the symbol appearance probability convergence determination means 106 described above, and has a symbol appearance probability calculation result P.
Determine whether _Tmp has converged.

B-3. Operation of Embodiment Next, the operation of this embodiment will be described. When an operator sets a document on the automatic document feeder 1 and inputs an instruction to start a copying operation from the user interface 14, the document set on the automatic document feeder 1 reaches the predetermined reading position of the image reading device 2. The image is conveyed, photoelectrically converted by the image reading device 2, and read as electrical image data. The read image data is supplied to the encoding processing unit 4.

In the encoding processing unit 4, first, the symbol appearance probability setting means 107 causes the arithmetic encoding processing means 102a to 102n to generate symbol appearance probability initial values.
P_Set_Init is set as the symbol appearance probability value P_Set. Next, the arithmetic coding processing means 102a to 102.
An image input request signal is output from n to the image dividing unit 101. When these image input request signals are supplied, the 103 divides the input image information into regions and supplies the divided image information to the arithmetic coding processing means 102a to 102n.

Next, each of the arithmetic coding processing means 102a-1.
In 02n, the symbol appearance probability initial value P_Set_Ini
The image information of each area is encoded based on t. After that, each of the arithmetic coding processing means 102a to 102n appropriately changes the symbol appearance probability value based on the result of the encoding processing of the image information received by each (however, the changed symbol appearance probability value is the outside). Is not output to, and is used exclusively for the encoding processing inside each arithmetic encoding processing means).
Then, when the encoding process for each corresponding region is finished, the output of the image input request signal to the image dividing unit 100 is stopped, and the encoding process end signal is sent to the encoding operation control unit 103 and the symbol appearance probability calculating unit 105. In addition to being output to Block 1-1, Block 2-1, ...
The appearance frequency information of each symbol obtained at the time of performing the coding processing of Block n-1 is the symbol appearance probability calculating means 10
It is output to 5.

Next, the symbol appearance probability calculation means 105
When the encoding process end signal output from the first arithmetic encoding processing unit 102a to the Nth arithmetic encoding processing unit 102n and the appearance frequency information of each symbol are all input, the first arithmetic encoding processing unit 102a. ~ Nth arithmetic coding processing means 1
In 02n, the appearance frequency of each symbol in all the image information that has already been processed is totaled, and the symbol appearance probability calculation result P_Tmp and the symbol appearance probability determination result P_Det are output. When the above processing is completed, the symbol appearance probability calculation means 105
Outputs the calculation result of the symbol appearance probability to the symbol appearance probability convergence determination means 106, and outputs the determination result of the symbol appearance probability value set in each arithmetic coding processing means to the symbol appearance probability setting means 107.

Next, the symbol appearance probability convergence determining means 10
In No. 6, it is determined whether or not the symbol appearance probability calculation result P_Tmp has converged. However, since the history of the symbol appearance probability calculation result P_Tmp does not exist at this time (initial state), it is said that it has not converged. To be judged. Therefore, the symbol appearance probability determination result P_Det is the symbol appearance probability value P_S.
As et, each arithmetic coding processing means 102a to 102n
Is set to Then, when the setting of the symbol appearance probability value P_Set is completed, the encoding processing start control means 104 outputs an encoding processing start signal to each of the arithmetic encoding processing means 102a to 102n. As a result, an image input request signal is output from each of the arithmetic coding processing means 102a to 102n to the image dividing means 101, and Block 1-2 and Block
The coding processing of 2-2, ..., Block n-2 is performed.

Next, the arithmetic coding processing means 102a-1.
02n outputs the coded data of these Blocks and the symbol appearance frequency information. The symbol appearance probability convergence determination means 106 determines whether or not the new symbol appearance frequency information satisfies the convergence condition with respect to the past aggregation result of the symbol appearance frequency information. If it is determined here that it has not converged, the symbol appearance probability determination result P
_Det is set as a symbol appearance probability value P_Set in each of the arithmetic coding processing means 102a to 102n. This process is repeated until the symbol appearance probability convergence determination means 106 determines that the symbol has converged.

Eventually, the symbol appearance probability convergence determining means 1
If it is determined in “06” that “converged”, the symbol appearance probability convergence determination means 106 causes the encoding operation control means 103.
To the symbol appearance probability convergence determination signal. When the symbol appearance probability convergence determination signal is input from the symbol appearance probability convergence determination means 106, the encoding operation control means 1
03 does not set the symbol appearance probability value in each arithmetic coding processing means, but the first arithmetic coding processing means 102a to 102a.
It instructs the encoding process start control unit 104 to activate the encoding process in the Nth arithmetic encoding process unit 102n.
The encoding process start control unit 104 outputs an encoding process start signal to the first arithmetic encoding process unit 102a to the Nth arithmetic encoding process unit 102n to instruct the start of the encoding process.

After the symbol appearance probability convergence determining means 106 determines that the calculation result of the appearance probability value of the symbol converges to the appearance probability value that matches the property of the input image information to be processed, the area [1] The image information of the area [N] is independently encoded by the first arithmetic encoding processing means 102a to the Nth arithmetic encoding processing means 102n, which are assigned to each area. At that time, the symbol appearance probability value P_Set used in each of the arithmetic coding processing means 102a to 102n is the same as each of the arithmetic coding processing means 10.
2a to 102n are appropriately changed according to the variation of the appearance probability of the symbols, which are generally different from each other.

When the image information is encoded in this way, the encoding operation control means 103 in the encoding processing unit 4 outputs an image transfer request to the bus arbiter 6. The CPU 16 is a DM that performs DMA transfer between the encoding processing unit 4 and the page memory 5.
The transfer start address is set in the channel 2 of the A controller 7.

Code data encoded by the encoding processing unit 4 by the arbitration of the bus arbiter 6 is written in the page memory 5 by DMA transfer. When the code data for one line or several lines is written in the page memory 5, it can be read out to the decoding processing unit 10, and the CPU 16 DMs the DMA transfer between the decoding processing unit 10 and the page memory 5.
The transfer start address is set in the channel 3 of the A controller 7, and the processing in the decoding processing unit 10 is started.
That is, the code data written in the page memory 5 is sent to the decoding processing unit 10 by DMA transfer by the arbitration of the bus arbiter 6 and restored to the original image data, and then recorded and output on a sheet by the image recording device 13. ..

Here, details of the operation of the decoding processing unit 10 will be described. When the decoding process is started in the decoding processing unit 10, the decoding operation control unit 109 first causes the symbol appearance probability setting unit 113 to perform the first arithmetic decoding processing unit 1.
08a to Nth arithmetic decoding processing means 108n are instructed to set the initial value of the symbol appearance probability, and first arithmetic decoding processing means 108a to Nth arithmetic decoding processing means 10 are provided.
Decryption process start control means 1 for activating decryption process at 8n
Give 10 instructions. The symbol appearance probability setting means 113
As the symbol appearance probability value P_Set, the symbol appearance probability initial value P_Set_Init is used as the first arithmetic decoding processing means 108a.
~ Set to the Nth arithmetic decoding processing means 108n.

When the setting of the initial value by the symbol appearance probability setting means 113 is completed, the decoding process start control means 11
0 outputs a decoding processing start signal to the first arithmetic decoding processing means 108a to the Nth arithmetic decoding processing means 108n to instruct to start the decoding processing. First arithmetic decoding processing means 108a, second activated by the decoding processing activation signal
The arithmetic decoding processing means 108b, ..., And the Nth arithmetic decoding processing means 108n each output a code data input request signal, and based on the symbol appearance probability initial value P_Set_Init set by the symbol appearance probability setting means 113. , Area [1], area [2], ..., Input image information (Block 1-1, Block 2-
, ..., Block n−1) corresponding to the encoded data is decoded. The arithmetic coding processing means 1 described above
As in the case of 02a to 102n, the symbol appearance probability value P_Set used inside each arithmetic decoding processing means 108a to 108n is appropriately changed according to the coded data received by each arithmetic coding processing means.

Each arithmetic decoding processing means is a Block
When the decoding process of the code data corresponding to 1-1, Block 2-1, ..., Block n-1 is completed, the output of the code data input request signal is stopped and the decoding process end signal is decoded. Control means 109, symbol appearance probability calculation means 111
Output to Further, each of the arithmetic decoding processing means 108a to 108n has a block 1-1, a block 2-1 and a block 2-1.
... At the time when the decoding process corresponding to Block n-1 is performed, the appearance frequency information of each symbol is output to the symbol appearance probability calculating unit 111.

The symbol appearance probability calculating means 111 has a first
When the decoding process end signal output from the arithmetic decoding processing unit 108a to the Nth arithmetic decoding processing unit 108n and the appearance frequency information of each symbol are all input, the first arithmetic decoding processing unit 108a to the Nth arithmetic operation. The appearance frequency of each symbol in all the image information that has already been processed by the decoding processing means 108n is totaled, and the symbol appearance probability calculation result P_Tmp and the symbol appearance probability determination result P_Det are output. When the above processing is completed, the symbol appearance probability calculation means 111 sends the symbol appearance probability calculation result P_Tmp to the symbol appearance probability convergence determination means 112, and the symbol appearance probability determination result P_Det.
Is supplied to the symbol appearance probability setting means 113.

The symbol appearance probability convergence determination means 112 is
In order to determine whether or not the symbol appearance probability calculation result P_Tmp converges to the appearance probability value that matches the property of the image information to be processed, the history of the symbol appearance probability calculation results input from the symbol appearance probability calculation unit 111. Is held, the amount of change in the symbol appearance probability during a certain period is determined, and when the amount of change in the symbol appearance probability during the certain period converges within a predetermined threshold, the appearance probability value of each symbol It is determined that the calculation result has converged to the appearance probability value that matches the property of the image information to be processed.

When the symbol appearance probability convergence determining means 112 determines that the calculation result of the symbol appearance probability value has not converged to the appearance probability value that matches the property of the image information to be processed, the decoding operation control is performed. The means 109 instructs the symbol appearance probability setting means 113 to set the symbol appearance probability value P_Set to the first arithmetic decoding processing means 108a to the Nth arithmetic decoding processing means 108n, and at the same time,
The decoding process start control unit 110 is instructed to start the decoding process in the arithmetic decoding process unit 108a to the Nth arithmetic decoding process unit 108n. Symbol appearance probability setting means 113
Sets the symbol appearance probability determination result P_Det input from the symbol appearance probability calculating means 111 in the first arithmetic decoding processing means 108a to the Nth arithmetic decoding processing means 108n. When the symbol appearance probability setting means 113 completes the setting of the symbol appearance probability value, the decoding processing start control means 110 sends the decoding processing start signal to the first arithmetic decoding processing means 108a to Nth arithmetic decoding processing means 108n. Is output to instruct the start of the decoding process.

The above processing is repeatedly executed until the calculation result of the appearance probability value of the symbol converges to the appearance probability value suitable for the property of the image information to be processed. When the symbol appearance probability convergence determination means 112 determines that the symbol appearance probability calculation result P_Tmp converges to the appearance probability value that matches the property of the image information to be processed, the symbol appearance probability convergence determination means 112 performs the decoding operation. Control means 1
09, a symbol appearance probability convergence determination signal is output. When the symbol appearance probability convergence determination signal is input from the symbol appearance probability convergence determination means 112, the decoding operation control means 109 does not set the symbol appearance probability value in each arithmetic decoding processing means, but performs the first arithmetic operation. Decoding processing means 10
8a to the decoding process start control unit 110 to start the decoding process in the Nth arithmetic decoding process unit 108n. The decoding processing start control means 110 outputs a decoding processing start signal to the first arithmetic decoding processing means 108a to Nth arithmetic decoding processing means 108n, and instructs the start of the decoding processing.

After the symbol appearance probability convergence determining means 112 determines that the calculation result of the appearance probability value of the symbol converges to the appearance probability value that matches the property of the input image information to be processed, the area [1] -The image information of the area [N] is independently decoded by the first arithmetic decoding processing means 108a to the N-th arithmetic decoding processing means 108n, which are assigned one for each area, and each arithmetic decoding is performed. The image information obtained as a result of the decoding processing by the processing means is
The images are combined by the image combining unit 114 and output as output image information.

The above processing is repeated for the number of pages of the document, and the copy of the first copy is completed. On the other hand, when the code data written in the page memory 5 corresponds to one page, the CPU
Reference numeral 16 sets a transfer start address in the channel 5 of the DMA controller 7 which performs the DMA transfer between the page memory 5 and the image storage device 8 and activates the writing process to the image storage device 8. The code data written in the page memory 5 is written in the image storage device 8 by DMA transfer by the arbitration of the bus arbiter 6. This is repeated for the number of pages of the original document during the copy period of the first copy. For copying the second and subsequent copies, the code data written in the image storage device 8 is transferred to the decoding processing unit 10 by DMA transfer, restored to the original image data, and then recorded and output on paper by the image recording device 13. .

C. Second Embodiment Next, a second embodiment of the present invention will be described. The overall configuration of the second embodiment is similar to that of the first embodiment (FIG. 1), but the encoding processing unit 4 and the decoding processing unit 10 are configured as shown in FIGS. 4 and 5, respectively. . The details of each will be described below. In these figures, parts corresponding to those in FIGS. 2 and 3 are designated by the same reference numerals, and the description thereof will be omitted.

(Encoding Processor 4) First, the encoding processor 4
The detailed configuration of will be described with reference to FIG. 121 in the figure
Is an image dividing means, and when the image information to be encoded is inputted through the selector 3, this image information is divided into a plurality of blocks as shown in FIG. Thereafter, in FIG. 8, Block 1-1, Block 1-2, ..., Block 1-m are areas [1], Block 2-1, Block 2-2, ..., Block 2-m are areas. [2], ::: Block n-1, Block n-2, ..., Block nm are referred to as a region [N]. The image dividing unit 121 sequentially outputs the image information of the corresponding area to each of the arithmetic coding processing units 102a to 102n in response to the image input request signal supplied from each of the arithmetic coding processing units 102a to 102n.

Next, reference numeral 123 is an encoding operation control means, which, similarly to the encoding operation control means 103 in the first embodiment, issues an appropriate command to other components in order to proceed with the encoding process. Output. However, in the present embodiment, the order in which the encoding processing unit activation signals are output to the arithmetic encoding processing means 102a to 102n is different from that in the first embodiment. That is, the encoding operation control means 123
First activates the first arithmetic coding processing means 102a,
Next, the first arithmetic coding processing means 102a and the second arithmetic coding processing means 102b are activated, and the number of arithmetic coding processing means activated in the same manner is increased by one until the number reaches "N". Then, an instruction is output to the encoding process start control means 104. Further, in the present embodiment, the one corresponding to the symbol appearance probability convergence determination means 106 of the first embodiment is not provided.

When the encoding process is started in the above configuration, the encoding operation control means 123 first causes the symbol appearance probability setting means 107 to the first arithmetic encoding processing means 102a.
Is set to the initial value of the symbol appearance probability, and the coding process start control unit 104 is instructed to start the coding process in the first arithmetic coding processing unit 102a. The symbol appearance probability setting means 107 sets the symbol appearance probability initial value P_Set_Init in the first arithmetic coding processing means 102a. When the setting of the initial value by the symbol appearance probability setting means 107 is completed, the encoding processing start control means 104 outputs an encoding processing start signal to the first arithmetic encoding processing means 102a to instruct the start of the encoding processing. .

In the first arithmetic encoding processing means 102a activated by the encoding processing activation signal, the image dividing means 12
The image input request signal is output to 1, and the predetermined number of pixels in the area [1] after the division by the image dividing unit 121 based on the symbol appearance probability initial value P_Set_Init set by the symbol appearance probability setting unit 107. The input image information (Block 1-1) is encoded by the arithmetic encoding method described above. That is, the first arithmetic coding processing means 102.
a The symbol appearance probability value P_Set used internally is the symbol appearance probability initial value P_Set_Init at the beginning of the process.
However, after that, it is appropriately updated based on the generation ratio of each encoded data. Then, when the encoding process of Block 1-1 is completed, the output of the image input request signal to the image dividing unit 121 is stopped, and the encoding process end signal is sent to the encoding operation control unit 123 and the symbol appearance probability calculating unit 105.
And the appearance frequency information of each symbol obtained at the time of performing the coding processing of Block 1-1 is output to the symbol appearance probability calculating unit 105.

The symbol appearance probability calculation means 105 has a first
When the encoding processing end signal output from the arithmetic encoding processing means 102a and the appearance frequency information of each symbol are input,
The appearance frequency of each symbol in the image information already processed by the first arithmetic coding processing means 102a is totaled, the appearance probability is calculated, and the symbol appearance set in each arithmetic coding processing means based on the calculation result. The probability determination result P_Det is determined and output to the symbol appearance probability setting means 107.

The encoding operation control means 123 receives the encoding processing end signal output from the first arithmetic encoding processing means 102a and inputs the encoding appearance probability setting means 107 to the first arithmetic encoding processing means 102a, The second arithmetic coding processing means 102b is instructed to set the symbol appearance probability, and the coding processing start control is performed to activate the coding processing in the first arithmetic coding processing means 102a and the second arithmetic coding processing means 102b. Instruct the means 104. The symbol appearance probability setting means 107 outputs the symbol appearance probability determination result P_Det input from the symbol appearance probability calculating means 105 to the first arithmetic coding processing means 102a and the second arithmetic coding processing means 10.
2b and the symbol appearance probability value P_Set.
When the setting of the symbol appearance probability value P_Set is completed, the encoding processing start control means 104 causes the first arithmetic encoding processing means 1 to operate.
02a and the second arithmetic coding processing means 102b, the coding processing start signal is output, and thereby the start of the coding processing is instructed.

In the first arithmetic coding processing means 102a activated by the encoding processing activation signal, the image dividing means 12 is used.
1, an image input request signal is output, and based on the symbol appearance probability value P_Set set by the symbol appearance probability setting unit 107, the image dividing unit 121 determines a predetermined next region in the region [1] after the division. The input image information (Block 1-2) of the number of pixels is encoded by the arithmetic encoding method. On the other hand, in parallel with the encoding processing by the first arithmetic encoding processing means 102a, the second arithmetic encoding processing means 102b outputs an image input request signal to the image dividing means 121 and the symbol appearance probability setting means. Based on the symbol appearance probability value P_Set set by 107, the image dividing means 121 encodes the input image information (Block 2-1) of a predetermined number of pixels in the area [2] after the division by the arithmetic encoding method. To process.

The first arithmetic coding processing means 102a and the second arithmetic coding processing means 102b are Block 1-2 and Block 2-
When the encoding process of No. 1 is completed, the output of the image input request signal to the image dividing unit 121 is stopped, and the encoding process end signal is output to the encoding operation control unit 123 and the symbol appearance probability calculating unit 105, respectively. At the same time, the appearance frequency information of each symbol obtained as a result of the encoding process is output to the symbol appearance probability calculating means 105.

When the coding process end signal and the appearance frequency information of each symbol output from the first arithmetic coding processing means 102a and the second arithmetic coding processing means 102b are input, the symbol appearance probability calculating means 105 The appearance frequency of each symbol is totaled, and the first arithmetic coding processing means 102a and the second arithmetic coding processing means 102a
The appearance probability of each symbol in all image information that has already been processed by the arithmetic coding processing means 102b is calculated, and the symbol appearance probability value set in each arithmetic coding processing means is determined based on the calculation result, and the symbol appearance It outputs to the probability setting means 107.

The coding operation control means 123 includes a first arithmetic coding processing means 102a and a second arithmetic coding processing means 102.
When the encoding process end signals output from the b and b are input, the symbol appearance probability setting unit 107 is input to the first arithmetic encoding processing unit 102a and the second arithmetic encoding processing unit 1.
02b, and instructs the third arithmetic coding processing means to be activated next to set the symbol appearance probability. Further, the encoding operation control means 123 activates the encoding processing in the first arithmetic encoding processing means 102a, the second arithmetic encoding processing means 102b, and the third arithmetic encoding processing means, and the encoding processing start control means 10
Instruct 4. The symbol appearance probability setting means 107 sets the symbol appearance probability value input from the symbol appearance probability calculating means 105 in the first arithmetic coding processing means 102a, the second arithmetic coding processing means 102b, and the third arithmetic coding processing means. To do. When the symbol appearance probability setting means 107 completes the setting of the symbol appearance probability values, the encoding processing start control means 104 causes the first arithmetic encoding processing means 102a and the second arithmetic encoding processing means 102a.
An encoding processing start signal is output to the arithmetic encoding processing means 102b and the third arithmetic encoding processing means to instruct the start of the encoding processing.

The above processing is performed by each arithmetic coding processing means 10.
This is repeated until all (N) arithmetic coding processing units 2a to 102n are activated. After the coding processing in all the arithmetic coding processing means is activated, one piece of image information in each of the areas [1] to [N] is assigned to each arithmetic coding processing means 102a. The encoding processing is performed in parallel by 102n.

(Decoding Processing Unit 10) Next, the detailed configuration of the decoding processing unit 10 will be described with reference to FIG. 1 in the figure
Reference numeral 29 denotes a decoding operation control means, which outputs a command as appropriate to other constituent elements in order to proceed with the decoding process, similarly to the decoding operation control means 109 in the first embodiment. However, in the present embodiment, the order in which the decoding processing unit activation signal is output to each of the arithmetic decoding processing means 108a to 108n is different from that in the first embodiment. That is, the decoding operation control means 129 first activates the first arithmetic decoding processing means 108a, then activates the first arithmetic decoding processing means 108a and the second arithmetic decoding processing means 108b, and so on. An instruction is output to the decoding processing start control means 110 so that the number of arithmetic decoding processing means activated by 1 increases by one until it reaches "N". Further, in the present embodiment, the one corresponding to the symbol appearance probability convergence determination means 112 of the first embodiment is not provided.

When the decoding process is started in the above configuration, the areas [1] to [N] input to the decoding processing unit are input.
Code data [1] to code data [N] of No. 1 according to the code data input request signals from the first arithmetic decoding processing means 108a to the Nth arithmetic decoding processing means 108n, which are assigned one for each area. , Is input to each arithmetic decoding processing means.

When the input of coded data to the decoding processing section is started, the decoding operation control means 129 first causes the symbol appearance probability setting means 113 to the first arithmetic decoding processing means 1.
08a is instructed to set the initial value of the symbol appearance probability, and the decoding process start control unit 110 is instructed to start the decoding process in the first arithmetic decoding process unit 108a.
The symbol appearance probability setting means 113 sets the symbol appearance probability initial value P_Set_Init to the first arithmetic decoding processing means 108a.
Set to. When the setting of the initial value by the symbol appearance probability setting means 113 is completed, the decoding process start control means 11
0 outputs a decoding processing start signal to the first arithmetic decoding processing means 108a to instruct the start of the decoding processing.

The first arithmetic decoding processing means 108a activated by the decoding processing activation signal outputs the code data input request signal and sets the symbol appearance probability value set by the symbol appearance probability setting means 113 as an initial value. When the code data corresponding to Block 1-1 in the area [1] is decoded and the decoding process of the code data corresponding to Block 1-1 is completed, the output of the code data input request signal is stopped and the decoding is performed. Decoding operation end signal decoding operation control means 12
9. The appearance frequency information of each symbol obtained at the time of decoding the code data corresponding to Block 1-1 is output to the symbol appearance probability calculating unit 111 while being output to the symbol appearance probability calculating unit 111. Output to.

The symbol appearance probability calculation means 111 has a first
When the decoding process end signal output from the arithmetic decoding processing unit 108a and the appearance frequency information of each symbol are input,
The appearance frequency of each symbol in the image information that has already been processed by the first arithmetic decoding processing means 108a is totaled, the appearance probability is calculated, and the symbol appearance set in each arithmetic decoding processing means based on the calculation result. The probability value is determined and output to the symbol appearance probability setting means 113.

When the decoding processing end signal output from the first arithmetic decoding processing means 108a is input, the decoding operation control means 129 causes the symbol appearance probability setting means 113 to output the first arithmetic decoding processing means 108a, Instructing the second arithmetic decoding processing means 108b to set the symbol appearance probability, and starting the decoding processing in the first arithmetic decoding processing means 108a and the second arithmetic decoding processing means 108b controls the decoding processing start. Instruct the means 110. The symbol appearance probability setting means 113 sets the symbol appearance probability value input from the symbol appearance probability calculating means 111 in the first arithmetic decoding processing means 108a and the second arithmetic decoding processing means 108b. When the symbol appearance probability setting unit 113 completes the setting of the symbol appearance probability value, the decoding process start control unit 110 sends a decoding process start signal to the first arithmetic decoding processing unit 108a and the second arithmetic decoding processing unit 108b. Is output to instruct the start of the decoding process.

The first arithmetic decoding processing means 108a activated by the decoding processing activation signal outputs the code data input request signal and sets the symbol appearance probability value set by the symbol appearance probability setting means 113 as an initial value. The code data corresponding to Block 1-2 of the next area in the area [1] is decoded by the arithmetic coding method. In parallel with the decoding processing by the first arithmetic decoding processing means 108a, the second
The arithmetic decoding processing means 108b outputs a code data input request signal, and sets the symbol appearance probability value set by the symbol appearance probability setting means 113 as an initial value to a code corresponding to Block 2-1 in the area [2]. The data is decoded by the arithmetic coding method.

The first arithmetic decoding processing means 108a and the second arithmetic decoding processing means 108b are Block 1-2 and Block 2-
When the decoding process of the code data corresponding to 1 is completed, the output of the code data input request signal is stopped, and the decoding process end signal is output to the decoding operation control unit 129 and the symbol appearance probability calculation unit 111. At the same time, the appearance frequency information of each symbol obtained as a result of the decoding process is output to the symbol appearance probability calculating unit 111.

When the decoding process end signal and the appearance frequency information of each symbol output from the first arithmetic decoding processing means 108a and the second arithmetic decoding processing means 108b are input, the symbol appearance probability calculating means 111 The appearance frequency of each symbol is totaled, and the first arithmetic decoding processing means 108a and the second arithmetic decoding processing means 108a
The appearance probability of each symbol in all the image information that has already been processed by the arithmetic decoding processing means 108b is calculated, the symbol appearance probability value set in each arithmetic decoding processing means is determined based on the calculation result, and the symbol appearance It outputs to the probability setting means 113.

The decoding operation control means 129 is composed of the first arithmetic decoding processing means 108a and the second arithmetic decoding processing means 108.
When the decoding process end signal output from b is input,
The symbol appearance probability setting means 113 is instructed to set the symbol appearance probability to the first arithmetic decoding processing means 108a, the second arithmetic decoding processing means 108b, and the third arithmetic decoding processing means to be activated next. Arithmetic decoding processing means 10
8a, the second arithmetic decoding processing means 108b, and the third arithmetic decoding processing means are instructed to the decoding processing start control means 110 to start the decoding processing. Symbol appearance probability setting means 11
3 is a first arithmetic decoding processing means 108 for the symbol appearance probability value input from the symbol appearance probability calculating means 111.
a, the second arithmetic decoding processing means 108b, and the third arithmetic decoding processing means. Symbol appearance probability setting means 113
When the setting of the symbol appearance probability value by is completed, the decoding processing start control means 110 causes the first arithmetic decoding processing means 10 to operate.
8a, the second arithmetic decoding processing means 108b, and the third arithmetic decoding processing means output a decoding processing start signal to instruct the start of the decoding processing.

The above processing is performed by the first arithmetic decoding processing means 1
08a to all of the Nth arithmetic decoding processing means 108n (N
It is repeated until the number of arithmetic decoding processing means is activated. After the decoding processing in all the arithmetic decoding processing means is activated, the code data [1] to the code data [N] corresponding to the area [1] to the area [N] are one for each area. The image information obtained as a result of being subjected to decoding processing in parallel by the first arithmetic decoding processing means 108a to the Nth arithmetic decoding processing means 108n, which are respectively allocated, and the decoding processing being performed by each arithmetic decoding processing means, The images are combined by the image combining unit 114 and output as output image information.

D. Third Embodiment Next, a third embodiment of the present invention will be described. The overall configuration of the third embodiment is similar to that of the first embodiment (FIG. 1), but the encoding processing unit 4 and the decoding processing unit 10 are configured as shown in FIGS. 6 and 7, respectively. . The details of each will be described below. In these figures, parts corresponding to those in FIGS. 2 and 3 are designated by the same reference numerals, and the description thereof will be omitted.

(Encoding Processor 4) First, the encoding processor 4
The detailed configuration of will be described with reference to FIG. In the figure 131
Is an image dividing means, and performs different operations depending on the level of an image dividing means control signal described later. That is, when the image dividing unit control signal is “0”, the image dividing unit 131 outputs all the encoding target image information (without division) to the first image information.
It is supplied to the arithmetic coding processing means 102a. On the other hand, when the image dividing means control signal becomes "1", the image dividing means 1
Reference numeral 31 divides the encoding target image information into a plurality of blocks, and outputs the blocks to the arithmetic encoding processing means 102a to 102n. When dividing the encoding target image information into a plurality of blocks, the block dividing method is the same as that of the second embodiment.

Therefore, in this embodiment, the encoding target image information is divided into a plurality of blocks as shown in FIG. Below, the image dividing means control signal is "0".
A block (Block A) in the state of is called an area [A]. In addition, blocks (Block 1-1, Block 1-2, ..., After the image dividing means control signal becomes “1”)
Block 1-m, ..., Block n-m) are referred to as region [1] to region [N] as in the second embodiment.

Next, reference numeral 133 denotes an encoding operation control means, which, similarly to the encoding operation control means 103 in the first embodiment, sends an appropriate command to other components in order to proceed with the encoding process. Output. However, in the present embodiment, the order in which the encoding processing unit activation signals are output to the arithmetic encoding processing means 102a to 102n is different from that in the first embodiment. That is, the encoding operation control unit 133
Is to output the image dividing means control signal of "0",
Only the first arithmetic coding processing means 102a is activated to code the image information of the area [A].

Further, when the image dividing means control signal is set to "1", the encoding operation control means 133 activates all the arithmetic encoding processing means 102a to 102n accordingly. Further, in the present embodiment, the symbol appearance frequency calculating means 105 determines the symbol appearance frequency information and the encoding processing end signal output by the first arithmetic encoding processing means 102a.
However, these signals output from other arithmetic coding processing means are not supplied to the symbol appearance probability calculating means 105. As a result, in the symbol appearance probability calculation means 105, the symbol appearance probability calculation result P_Tmp and the symbol appearance probability determination result P_Det are determined based only on the encoding processing result of the first arithmetic encoding processing means 102a. It will be.

Further, the symbol appearance probability convergence determining means 10
6 is the symbol appearance probability calculation result, as in the first embodiment.
When the history of P_Tmp is stored and a new symbol appearance probability calculation result P_Tmp is supplied, this is compared with the previously supplied symbol appearance probability calculation result P_Tmp, and whether the difference between the two matches the predetermined convergence condition. When it is determined whether or not it is “matched”, a symbol appearance probability convergence determination signal is output. When receiving the symbol appearance probability convergence determination signal, the encoding operation control means 133 sets the image division means control signal to "1" and activates all the arithmetic encoding processing means 102a to 102n.

When the encoding process is started in the above configuration, the encoding operation control means 133 first causes the first arithmetic encoding processing means 102a to send the symbol appearance probability initial value P_Set_I.
Symbol appearance probability setting means 10 so as to set nit
7, and the first arithmetic coding processing means 10
The encoding process start control means 104 is instructed to start the activation of the encoding process in 2a. When these instructions are given, the symbol appearance probability setting means 107 sets the symbol appearance probability initial value P_Set_Init in the first arithmetic coding processing means 102a. When this setting is completed, the encoding processing start control means 104 causes the first arithmetic encoding processing means 102.
A coding process start signal is output to a to instruct the start of the coding process.

The encoding target image information input to the encoding processing section is input to the image dividing means 131. At this time, the encoding operation control unit 133 instructs the image division unit 131 not to perform the image division process by setting the image division unit control signal to “0”. Therefore, the image dividing unit 131 supplies all the supplied image information to the first arithmetic coding processing unit 102a. That is, the first arithmetic coding processing means 102a appropriately sends the image input request signal to the image dividing means 1 as the coding processing progresses.
31 and the image division means 131 sequentially outputs the image information accordingly.

The first arithmetic encoding processing means 102a activated by the encoding processing activation signal encodes the input image information based on the symbol appearance probability initial value P_Set_Init set by the symbol appearance probability setting means 107. . When the encoding process of the input image information of the predetermined number of pixels is completed, the output of the image input request signal to the image dividing unit 131 is stopped, and the encoding process end signal is transmitted to the encoding operation control unit 133 and the symbol appearance probability calculating unit. In addition to outputting to 105, the appearance frequency information of each symbol obtained at the time of coding the input image information of a predetermined number of pixels is output to the symbol appearance probability calculating means 105.

The symbol appearance probability calculation means 105 has a first
When the encoding processing end signal output from the arithmetic encoding processing means 102a and the appearance frequency information of each symbol are input,
The appearance frequency of each symbol in all the image information that has already been processed by the first arithmetic coding processing unit 102a is totaled, and the symbol appearance probability calculation result P_Tmp and the symbol appearance probability determination result P_D are obtained.
Output et and. Symbol appearance probability convergence determination means 10
In 6, as in the case of the first embodiment, it is determined whether or not the symbol appearance probability calculation result P_Tmp has converged.

When it is determined that "convergence has not occurred", the coding operation control means 133 codes the symbol appearance probability setting means 107 to start the coding processing in the first arithmetic coding processing means 102a. The conversion processing start control means 104 is instructed. The encoding processing start control means 104 outputs an encoding processing start signal to the first arithmetic encoding processing means 102a to instruct to start the encoding processing. As a result, the image input request signal is output again from the first arithmetic coding processing means 102a, and in response to this, the next image information is output from the image dividing means 131 to the first arithmetic coding processing means 102a.
Hereinafter, the same operation is repeated.

Thereafter, the symbol appearance probability convergence determination means 1
If it is determined in "06" that "converged", the symbol appearance probability convergence determination means 106 is the encoding operation control means 13.
For 3, the symbol appearance probability convergence determination signal is output.
When the symbol appearance probability convergence determination signal is input from the symbol appearance probability convergence determination means 106, the encoding operation control means 133 sets the image division means control signal to the “1” level, whereby the image division means 131 is set. Command to start the image division processing.

Further, the encoding operation control means 133 supplies the symbol appearance probability setting instruction signal to the symbol appearance probability setting means 107. As a result, the symbol appearance probability setting unit 107 has the arithmetic coding processing units 102a to 102a.
As the symbol appearance probability value P_Set of 102n, the symbol appearance probability determination result P_Det output from the symbol appearance probability calculating unit 105 is set. Symbol appearance probability value
When the setting of P_Set is completed, the encoding operation control means 133
Instructs the coding process start control unit 104 to start the coding process in each of the arithmetic coding processing units 102a to 102n.

After that, the image information of each area [1] to area [N] is calculated according to the image input request signal from each arithmetic coding processing means 102a to 102n assigned to each area. It is input to the encoding processing means. Each of the arithmetic coding processing means 102a to 102n uses the symbol appearance probability value P_Set set by the symbol appearance probability setting means 107 as an initial value, and inputs the input image information of the area [1] to area [N] by the arithmetic coding. Depending on the method, encoding processing is performed in parallel. Similarly, the areas [1] to [N] divided by the image dividing means 131 are thereafter.
The image information of 1 is independently encoded by each of the arithmetic encoding processing means 102a to 102n assigned to each area.

(Decoding Processing Unit 10) Next, the detailed configuration of the decoding processing unit 10 will be described with reference to FIG. 1 in the figure
Reference numeral 39 denotes a decoding operation control means, which outputs a command as appropriate to other constituent elements in order to proceed with the decoding process, similarly to the decoding operation control means 109 in the first embodiment. However, in the present embodiment, the order in which the decoding processing unit activation signal is output to each of the arithmetic decoding processing means 108a to 108n is different from that in the first embodiment.

That is, the decryption operation control means 139
First, the first arithmetic decoding processing means 108a is activated, and the area
After the encoded data corresponding to [A] is decoded, each of the arithmetic decoding processing means 108a to 108n is activated to operate them in parallel. Further, in the present embodiment, the symbol appearance probability calculation means 111 calculates the symbol appearance probability based on the decoding process end signal and the symbol appearance frequency information output from the first arithmetic decoding processing means 108a. P_Tmp and symbol appearance probability determination result P_Det are output.

When the decoding process is started in the above configuration, first, the code data [A] corresponding to the area [A] in FIG. 10 is input to the decoding processing unit. The input coded data [A] of the area [A] has the first coded data according to the coded data input request signal from the first arithmetic decoding processing means 108a.
It is input to the arithmetic decoding processing means 108a. When the input of coded data to the decoding processing unit is started, the decoding operation control unit 139 performs the combining process of the image data after the decoding process on the image combining unit 114 according to the image combining unit control signal. Instruct to output without output. At the same time, the decoding operation control means 139 makes the symbol appearance probability setting means 1
13 is instructed to the first arithmetic decoding processing means 108a to set the initial value of the symbol appearance probability, and to the decoding processing start control means 110 to start the decoding processing in the first arithmetic decoding processing means 108a. To do.

Further, the symbol appearance probability setting means 113.
Sets the symbol appearance probability initial value P_Set_Init in the first arithmetic decoding processing means 108a. When the setting of the initial value by the symbol appearance probability setting means 113 is completed, the decoding processing start control means 110 causes the first arithmetic decoding processing means 10 to operate.
8a is output to the decoding process start signal to instruct the start of the decoding process. The first arithmetic decoding processing means 108a activated by the decoding processing activation signal outputs the code data input request signal and outputs the symbol appearance probability initial value P_Set_In.
The code data corresponding to the predetermined number of pixels in the area [A] is decoded based on it, and when the code data corresponding to the predetermined number of pixels is decoded, the output of the code data input request signal is stopped. Then, the decoding process end signal is output to the decoding operation control unit 139 and the symbol appearance probability calculation unit 111, and each symbol obtained at the time when the decoding process of the code data corresponding to the predetermined number of pixels is performed. The appearance frequency information of is output to the symbol appearance probability calculating unit 111.

The symbol appearance probability calculation means 111 has a first
When the decoding process end signal output from the arithmetic decoding processing unit 108a and the appearance frequency information of each symbol are input,
The appearance frequency of each symbol in all the image information that has already been processed by the first arithmetic decoding processing means 108a is totaled, and the appearance probability value of each symbol is calculated. The symbol appearance probability calculation means 111 outputs the calculation result of the symbol appearance probability to the symbol appearance probability convergence determination means 112.

The symbol appearance probability convergence determination means 112 is
In order to determine whether or not the calculation result of the appearance probability value of each symbol input from the symbol appearance probability calculation means 111 has converged to the appearance probability value that matches the property of the image information to be processed, the symbol appearance probability calculation means The history of the calculation result of the symbol appearance probability input from 111 is held,
Obtaining the amount of change in the symbol appearance probability during a certain period, when the amount of change in the symbol appearance probability during the certain period converges within a predetermined threshold value, the calculation result of the appearance probability value of each symbol is the processing target. It is determined that the value has converged to the appearance probability value that matches the property of the image information.

When the symbol appearance probability convergence determining means 112 determines that the calculation result of the symbol appearance probability value has not converged to the appearance probability value that matches the property of the image information to be processed, the decoding operation control is performed. The means 139 instructs the symbol appearance probability setting means 113 to the decoding processing start control means 110 to start the decoding processing in the first arithmetic decoding processing means 108a. Decoding process start control means 110
Outputs a decoding process start signal to the first arithmetic decoding process means 108a to instruct the start of the decoding process. The above processing is repeatedly executed until the calculation result of the appearance probability value of the symbol converges to the appearance probability value that matches the property of the image information to be processed.

When the symbol appearance probability convergence determining means 112 determines that the calculation result of the symbol appearance probability value has converged to the appearance probability value that matches the property of the image information to be processed, the symbol appearance probability convergence determining means. 112 outputs a symbol appearance probability convergence determination signal to the decoding operation control means 139. When the symbol appearance probability convergence determination signal is input from the symbol appearance probability convergence determination means 112, the decoding operation control means 139 causes the code data corresponding to the areas [1] to [N] shown in FIG. First arithmetic decoding processing means 108a assigned one by one
-Instructing parallel decoding processing by the Nth arithmetic decoding processing means 108n, and starting the synthesizing processing of the image information after the decoding processing to the image synthesizing means 114 by the image synthesizing means control signal. Instruct.

The decoding operation control means 139 activates the image synthesizing process in the image synthesizing means 114, and at the same time, the calculation result of the symbol appearance probability value converges to the appearance probability value suitable for the property of the input image information to be processed. When it is determined that the symbol appearance probability P_Det is set to the first
The symbol appearance probability setting unit 1 is set to be set in the arithmetic decoding processing unit 108a to the Nth arithmetic decoding processing unit 108n.
Instruct 13. In response to this, the symbol appearance probability setting means 113 sets the symbol appearance probability determination result P_Det as the symbol appearance probability value P_Set in each of the arithmetic decoding processing means 108a to 108n.

When the symbol appearance probability setting means 113 completes the setting of the symbol appearance probability value P_Set, the decoding operation control means 139 causes the first arithmetic decoding processing means 108.
a to instruct the decoding process start control unit 110 to start the decoding process in the Nth arithmetic decoding process unit 108n. The decoding processing start control means 110 is the first arithmetic decoding processing means 1.
08a to Nth arithmetic decoding processing means 108n outputs a decoding processing start signal to instruct to start the decoding processing.

The first arithmetic decoding processing means 108a to the Nth arithmetic decoding processing means 108n activated by the decoding processing activation signal respectively output the code data input request signal and are set by the symbol appearance probability setting means 113. Using the symbol appearance probability value as an initial value, the code data [1] to code data [N] corresponding to the area [1] to area [N] are decoded in parallel by the arithmetic coding method.

Thereafter, the code data [1] to code data [N] corresponding to the areas [1] to [N] are set to 1 for each area.
First arithmetic decoding processing means 108a to which the individual pieces are assigned
The Nth arithmetic decoding processing means 108n independently perform the decoding processing. First arithmetic decoding processing means 108a
Image information corresponding to the area [A] and area [1] to area [N] obtained as a result of the decoding processing by the Nth arithmetic decoding processing means 108n is shown in FIG. It is combined into the form of one image shown and output as output image information.

E. Effects of the Embodiment According to the first embodiment, each arithmetic coding / decoding processing means has already been based on the appearance frequency information of each symbol obtained at the time when the coding processing of a predetermined number of pixels is completed. The appearance probability of each symbol in the entire processed image information is calculated, and the calculation result is set as the symbol appearance probability value used in the plurality of arithmetic encoding / decoding processing means, and the encoding / decoding of the next area is performed. By repeating the process of performing the
Compared to the case where each arithmetic encoding / decoding processing means is operated independently, the initial value of the appearance probability of each symbol set at the start of processing in each arithmetic encoding / decoding processing means It is possible to reduce the number of pixels to be processed until convergence to the appearance probability suitable for the property of the input image information, and it is possible to improve the coding efficiency.

Further, according to the second embodiment, control for shifting the processing start time of each arithmetic coding / decoding processing means is performed,
In M (<N) arithmetic coding / decoding processing means that have already been activated, the processing has already been performed based on the appearance frequency information of each symbol obtained at the time when the encoding processing of a predetermined number of pixels is completed. , The appearance probability of each symbol in the entire image information is calculated, and the calculation result is sent to the M arithmetic coding / decoding processing means and the (M + 1) th arithmetic coding / decoding processing means to be activated next. Is set as a symbol appearance probability value to be used for each, and the process of performing the encoding / decoding process of the next area is repeated, so that each arithmetic encoding / decoding process means is operated independently. To reduce the number of processing pixels until the initial value of the appearance probability of each symbol set at the start of processing in the arithmetic coding / decoding processing means converges to the appearance probability suitable for the property of the input image information to be processed. Can Effect is obtained that it becomes possible to improve the Goka efficiency. Further, according to the second embodiment, by performing control so as to shift the processing start time of each arithmetic coding / decoding means, it is possible to perform real-time input such that the scanner sequentially inputs pixels to the image processing apparatus. It is possible to obtain an effect that it is possible to enhance adaptability to a system that requires the processing of.

Further, according to the third embodiment, the specific arithmetic coding / decoding processing means among the plurality of arithmetic coding / decoding processing means ends the coding / decoding processing of a predetermined number of pixels. The appearance probability of each symbol in the entire image information that has already been processed is calculated based on the cumulative result of the appearance frequency of each symbol at the time, and the calculation result of the symbol appearance probability value is the property of the input image information to be processed. The specific arithmetic encoding / decoding processing means performs the encoding / decoding processing until the appearance probability value is adapted to the symbol occurrence probability value, and the symbol appearance probability value at the time when the calculation result of the symbol appearance probability value converges The arithmetic coding / decoding processing means is set as an initial value of the symbol appearance probability value, and the remaining area is coded / decoded to operate each arithmetic coding / decoding processing means independently. When compared to Then, the number of processing pixels until the initial value of the appearance probability of each symbol set at the start of processing in each arithmetic encoding / decoding processing unit converges to the appearance probability that matches the property of the input image information to be processed. Can be reduced to the same level as in the case where one piece of image information is encoded by one arithmetic encoding processing unit, and the encoding efficiency can be improved.

Specifically, the initial value of the appearance probability of each symbol set in the arithmetic coding processing means when the whole one piece of image information is coded by one arithmetic coding processing means is , Letting X pixels be the number of pixels to be processed until they converge to an appearance probability that matches the property of the input image information to be processed, a plurality (N) of arithmetic coding / decoding processing means are operated independently. In this case, the initial value of the appearance probability of each symbol set in each arithmetic coding processing unit at the start of processing converges to the appearance probability suitable for the property of the input image information to be processed (X × N) pixels. However, according to the first and second embodiments of the present invention,
It is possible to suppress an increase in the number of pixels until the initial value of the appearance probability of each symbol set at the start of processing in each arithmetic coding processing unit converges to the appearance probability that matches the property of the input image information to be processed. ((X + α) pixels) According to the third embodiment, the initial value of the appearance probability of each symbol set in each arithmetic coding processing unit at the start of processing appears in accordance with the property of the input image information to be processed. Since there is almost no increase in the number of pixels until the probability is converged, it is possible to minimize the decrease in coding efficiency.

F. Modifications It is needless to say that the present invention is not limited to the above-described embodiments, and various modifications can be made. In each of the above embodiments, the arithmetic coding method is used as an example of the learning algorithm, but it goes without saying that the learning algorithm is not limited to the arithmetic coding method. For example, a plurality of types of compressors for compressing image information to be encoded and a switch for selecting these compressors are provided, and encoding is performed to automatically select an optimum compressor based on the appearance pattern of past encoded data. Although a processing circuit is known, the present invention can be applied to such an encoding processing circuit.
That is, in place of the arithmetic coding processing means 102a to 102n in each of the above embodiments, it is preferable to provide a plurality of coding processing circuits for automatically selecting a compressor. In such a case, the "compressor type" corresponds to the "compression parameter" in the claims.

In the above-described first embodiment, the symbol appearance probability convergence determination means 106 (or 112) determines that "the symbol appearance probability calculation result P_Tmp has converged", and the symbol appearance probability determination result P_Det at that time is determined.
Is set as the symbol appearance probability value P_Set of each of the arithmetic coding processing means 102a to 102n (or each of the arithmetic decoding processing means 108a to 108n), thereafter, the symbol appearance probability setting means 107 sets the symbol appearance probability value.
P_Set was never reset.

However, even after that, the symbol appearance probability calculation means 105 (or 111) continues to calculate the symbol appearance probability calculation result P_Tmp, and the symbol appearance probability convergence determination means 106 (or 112) determines whether or not The first embodiment may be configured such that the variation amount of the symbol appearance probability calculation result P_Tmp is continuously calculated and the symbol appearance probability value P_Set is updated when the variation amount is large.

[0116]

As described above, claim 1 or 3
According to the configuration described above, the learning results of the plurality of data compression units are aggregated and then shared, and according to the configuration of the second structure, the learning results of the first data compression unit are transmitted to the second data compression unit. Since it can be used, it is possible to compress data at high speed while maintaining a high compression rate.

[Brief description of drawings]

FIG. 1 is a block diagram of a copying machine according to a first embodiment of the present invention.

FIG. 2 is a block diagram of an encoding processing unit 4 in the first embodiment.

FIG. 3 is a block diagram of a decoding processing unit 10 in the first embodiment.

FIG. 4 is an encoding processing unit 4 according to the second embodiment of the present invention.
It is a block diagram of.

FIG. 5 is a block diagram of a decoding processing unit 10 in the second embodiment.

FIG. 6 is an encoding processing unit 4 according to the third embodiment of the present invention.
It is a block diagram of.

FIG. 7 is a block diagram of a decoding processing unit 10 in a third embodiment.

FIG. 8 is an operation explanatory diagram of the first embodiment.

FIG. 9 is an operation explanatory diagram of the second embodiment.

FIG. 10 is an operation explanatory diagram of the third embodiment.

FIG. 11 is an explanatory diagram of an encoding process in the first to third embodiments.

FIG. 12 is an operation explanatory diagram of the symbol appearance probability convergence determination means 106.

FIG. 13 is an operation explanatory diagram of a conventional technique.

[Explanation of symbols]

102a to 102n Arithmetic encoding processing means (data compression means) 101, 121 Image division means (data division means) 105 Symbol appearance probability calculation means (aggregation means, appearance probability calculation means) 103 Encoding operation control means (compression parameter setting means) ) 107 symbol appearance probability setting means (compression parameter setting means) 123 encoding operation control means (starting control means) 106 symbol appearance probability convergence determination means (convergence condition determination means)

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Internal reference number FI technical display area H04N 1/413 D

Claims (3)

[Claims] 1. A plurality of data compression means for compressing and outputting input data while appropriately changing compression parameters based on a predetermined learning algorithm, and when the compression target data is input, the compression target data is input. A data dividing unit that divides the data and supplies the data to each of the data compressing units; an aggregate unit that aggregates learning results of the data compressing units; and a common unit for each data compressing unit based on the aggregate result of the aggregate unit And a compression parameter setting means for setting the compression parameter of the data compression apparatus. 2. The first and second data compression means for compressing and outputting the input data while appropriately changing the compression parameter based on a predetermined learning algorithm, and when the data to be compressed is input, The data dividing means for dividing the data to be compressed and supplying the divided data to the first and second data compressing means and the first data compressing means are activated, and then the second data is compressed.
And a compression parameter setting means for setting the initial value of the compression parameter in the second data compression means based on the learning result in the first data compression means. A data compression device characterized by the above. 3. The compression target data is composed of a plurality of types of symbols, and each of the data compression means converts a symbol having a high appearance probability into binary data having a short bit length, and also has an appearance probability. Of the low-symbol is converted into binary data having a long bit length to compress the supplied data, and an appearance probability calculating means for sequentially calculating the appearance probability of the symbols included in the compression target data, and the appearance And a convergence condition determining means for determining whether or not the variation amount of the probability satisfies a predetermined convergence condition, and stopping the operation of the compression parameter setting means when the convergence condition is satisfied. The data compression device according to claim 1.