JP2010136417A - Data compression device and data decompression device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a high-speed and high compression rate data compression device and a data decompression device regarding a compression device and a decompression device for various data such as character data, image data, etc. <P>SOLUTION: The compression device and the decompression device for probability-statistical encoded data on which variable length coding is performed in accordance with an appearance frequency, in which, in an order starting from the maximum appearance frequency, characters are represented with different numbers of bits and registered in distinguishable tables so as to be easily specified, thereby performing compression and decompression at high speed and with high efficiency. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a compression device and a restoration device for various data such as character data and image data.

  With the rapid progress of today's computers, various types of information such as character information, vector information, and image information are handled by computers, and the amount of data handled is rapidly increasing. Under such circumstances, in order to shorten the transmission time and to efficiently use the storage capacity of the storage device, a data compression method for compressing the data amount by omitting redundant portions included in the data and a restoration method thereof have been proposed. ing.

  Further, as an encoding method used when compressing data, a universal encoding method applicable to various data such as character information, vector information, and image information is adopted. Furthermore, universal encoding includes a dictionary-type encoding method that uses character string similarity and a probability statistical encoding method that uses character appearance frequency.

  A typical example of a dictionary-type encoding method is Lempel-Ziv encoding. In this encoding, two algorithms of a slide dictionary type (universal type) and a dynamic dictionary type (incremental decomposition type) have been proposed. ing. As an improvement of the slide dictionary type algorithm, LZSS encoding, QIC-122 encoding, which is a standard compression method for 1/4 inch cartridge magnetic tape, and the like are known. On the other hand, LZW (Lempel-Ziv-Welch) coding or the like is known as an improvement of the dynamic dictionary type algorithm.

  Further, the probability statistical compression method is a method for achieving a compression effect by assigning a short code length to characters having a high appearance probability according to the statistical appearance frequency (appearance probability) of each character. As a representative example of this probability statistical coding method, for example, an arithmetic coding method and a Huffman coding method are known. In the Huffman coding method, a code having a code length inversely proportional to the appearance frequency of a character (Huffman code) is used as a code for each character.

  On the other hand, there is also a composite data compression method having both the dictionary coding method and the probability statistical compression method, such as the LZH method for compressing the data compressed by the LZSS method by the Huffman method. The LZSS method is a method of compressing a character string as a unit, and the Huffman method is a method of compressing one character unit, and can be expected to complement each other.

  Here, a conventional slide dictionary type algorithm (LZ1) will be described. This algorithm is a method that requires a large amount of computation but can provide a high compression rate. That is, the encoded data is divided into a longest sequence that matches from an arbitrary position in the past data series (partial string), and encoded as a copy of the past character string. FIG. 27 is a diagram for explaining this principle.

  The P buffer shown in the figure stores encoded input data, and the pre-encoded data is input to the Q buffer. In this state, the character string input to the Q buffer is sequentially compared with the character string stored in the P buffer to obtain the longest character string that matches in the P buffer, and the corresponding longest character string in the P buffer. If exists, an encoding process is performed. For example, as shown in the figure, when there is a matching character string in the P buffer, [start position p1 of the longest matching sequence (partial character string) in the P buffer], [maximum matching in the Q buffer] The compressed data is generated as the length q1]. By this encoding process, for example, the document information of the character code can be compressed to about ½. Moreover, there exists patent document 1 as invention which used the said slide dictionary type | mold algorithm.

  Next, probability statistical coding will be described. The probability statistical coding method is composed of an input buffer, a statistical model section, a frequency table, and an entropy coding section, as shown in FIG. 28. The statistical model section first scans all character strings and the appearance of each character. The frequency is calculated, and the entropy encoding unit assigns a code created based on the appearance probability obtained by the statistical model unit to each character.

  The appearance frequency obtained in the statistical model section includes a static encoding method in which the occurrence probability of each character is obtained in advance, a semi-adaptive encoding method that first obtains the occurrence probability of each character by scanning all character strings, and each character Is classified into an adaptive coding system that takes the frequency and recalculates the probability of occurrence each time. Incidentally, Patent Document 2 is disclosed as an invention using the probability statistical coding system.

Japanese Patent No. 3241788 Japanese Patent No. 3276860

However, the above compression method has the following problems.
First, in the dictionary-type encoding method, as described above, a character substring that matches the maximum length of the character string to be encoded is searched from the encoded character string, and the character substring is encoded as a copy. And a high data compression rate can be realized. In order to implement such an algorithm, it is necessary to adopt a configuration in which the data compression rate is further increased, and it is necessary to make the encoded data easy to use.

  For example, in order to increase the data compression rate in the above example, it is necessary to increase the number of characters stored in the P buffer and the number of characters stored in the Q buffer. However, if the number of characters stored in the P buffer or Q buffer is increased, the encoded data is not a multiple of 8 bits. In this case, troublesome processing such as bit padding is required when transferring data, resulting in a data compression method with extremely low processing efficiency.

  Also, when implemented in hardware, the window size is larger, which is a problem when the compression and decompression circuits are arranged on a single integrated circuit. That is, the circuit becomes enormous and the required hardware cost becomes very high. Further, when the window size is increased, the amount of comparison calculation becomes enormous and causes performance degradation.

  On the other hand, in the probability statistical encoding method, when one character is encoded, a code composed of integer bits is generated. On the other hand, in arithmetic coding, fractional bits can be assigned to one character. In arithmetic coding, the interval from 0 to less than 1 is sequentially narrowed according to the appearance frequency of each character constituting the data to be encoded. When processing for all characters is completed, a numerical value representing one point in the narrowed section is output as a code.

  Therefore, in the stochastic statistical coding, the processing speed is low because the coding is performed in two passes of the statistical model unit and the entropy coding unit. Also, because of variable length coding, it is necessary to perform bit operations in software, and this processing takes time.

  Therefore, the present invention provides a data compression and decompression device that is faster and has a higher compression efficiency by using a data compression device of probability statistical coding.

  According to the present invention, according to the present invention, in the data compression apparatus of probability statistical coding that performs variable length coding according to the appearance frequency, information for extracting data having the highest appearance frequency is stored in the first table and the second table. Information is recorded in 2 bits in the table, and the next two pieces of data are extracted in 4 bits in the first to third tables and the fifth table, and the next four pieces of data are extracted. Is recorded in 6 bits in the first to fourth tables and the sixth table, and information for extracting the next 16 data is recorded in the first to fourth tables and the seventh table. To provide a data compression apparatus having means for recording 8 bits and recording other data in the eighth table in 9 bits and means for recording data codes corresponding to the ranks of the appearance frequencies in the code table By It can be achieved Te.

With this configuration, data compression processing can be performed also by a data compression apparatus using probability statistical coding.
If only the most frequently appearing data and other data are included, information indicating whether the data is compressed data is recorded in the first table, and the code of the most frequently occurring data is recorded in the code table. Is recorded.

  Further, according to the present invention, in the restoration apparatus for probability statistical type encoded data that is variable length encoded according to the appearance frequency, the above-described problem is information regarding whether or not the data is compressed data with reference to the first table. If the data is not compressed data, the code data recorded in the eighth table is read. If the data is compressed data, the second table is referenced to obtain information on whether or not the data has the highest frequency of appearance. If the data has the highest frequency of appearance, means for reading out the code data of the corresponding frequency of appearance from the code table, and if the data does not have the highest frequency of appearance, the third table is further referred to Information about whether or not there are two more frequently-occurring data, and if the two most frequently appearing data, the fifth table is referred to and the data that has the second most frequently appearing data. Or Means for reading out the information indicating whether the data is the second largest number of data, and reading out the code data of the corresponding appearance frequency from the code table; and, if the two occurrence data are not the second most frequent data, further refer to the fourth table Then, information on whether or not there are four pieces of data having the next highest appearance frequency is acquired, and if the four pieces of data having the next highest occurrence frequency, the sixth table is further referred to, and the appearance frequency is 4 Means for obtaining information on whether the data is the seventh to seventh most frequent data, and reading out the code data of the corresponding appearance frequency from the code table; There is provided a data restoration device having means for referring to a table, obtaining information of 16 pieces of data having the next highest appearance frequency, and reading out code data of the corresponding appearance frequency from the code table. It can be achieved by the.

With this configuration, it is possible to realize a probability statistical type encoded data restoration apparatus.
If only the most frequently appearing data and other data are present, the first table is referred to obtain information on whether or not the data is compressed data. If the data is not compressed data, the corresponding code is read from the code table. This is a configuration for reading data.

  ADVANTAGE OF THE INVENTION According to this invention, the data compression and decompression | restoration apparatus of higher speed and high compression efficiency can be provided by the data compression apparatus of a stochastic statistical type encoding.

  In addition, the data compressed by the data compressing apparatus of the probability statistical coding can be restored at high speed and with high efficiency.

1 is a system configuration diagram of a data compression apparatus according to Embodiment 1. FIG. (A) is a figure explaining the structure of input data, (b) is a figure which shows the specific structure of compressed data, (c) And (d) demonstrates the structure of flag data. (E) is a figure explaining the structure of uncompressed data, (f) is a figure explaining the structure of compressed data. 6 is a diagram illustrating a configuration of Embodiment 2. FIG. It is a figure explaining the compression process of Embodiment 3. FIG. It is a figure which shows the example of identification data. It is a figure explaining the compression process of Embodiment 4. FIG. It is a figure which shows the example of identification data. FIG. 10 is a diagram illustrating compression processing according to a fifth embodiment. It is a figure explaining the structure of a preparation process part. It is a flowchart explaining the process which a preparation process part performs. It is a figure which shows the example of a compression format. It is a flowchart explaining the process which an entropy encoding process part performs. FIG. 10 is a diagram illustrating compression processing according to a sixth embodiment. It is a figure which shows the example of a determination pattern. It is a figure which shows the output format of the compression format which a compression efficiency determination part determines and outputs from a calculation result. It is a figure which shows the structure of a compression format. It is a flowchart which shows the process which an entropy encoding process part performs. FIG. 10 is a diagram for explaining an eighth embodiment. (A) is a figure explaining the structure of compressed data, (c) and (d) are figures explaining the structure of flag data, (e) is a figure explaining the structure of non-compressed data. (F) is a figure explaining the structure of compression data, (g) is a figure explaining the identification information of continuous length compression. It is a figure explaining Embodiment 9. FIG. It is a figure explaining Embodiment 10. FIG. It is a figure explaining the data decompression | restoration apparatus of Embodiment 11. FIG. It is a figure explaining the data decompression | restoration apparatus of Embodiment 12. 18 is a flowchart illustrating a data restoration device according to a thirteenth embodiment. FIG. 20 is a diagram illustrating data restoration processing according to the fourteenth embodiment. FIG. 20 is a diagram illustrating data restoration processing according to the fifteenth embodiment. It is a figure explaining the data compression process of a prior art example. It is a figure explaining the data compression process of a prior art example.

Embodiments of the present invention will be described below with reference to the drawings.
(Embodiment 1)
FIG. 1 is a system configuration diagram of a data compression apparatus according to the present embodiment.

  In the figure, compression means 1 is a first compression means using an adaptive dictionary-type coding method (Lempel-Ziv method), and compression means 2 is a semi-adaptive probability statistical coding. It is the 2nd compression means using the method. The input data string is input to the compression unit 1 and encoded by the compression unit 1. The data encoded by the compression means 1 is supplied to an intermediate data buffer (RAM) 13 and further compressed by the compression means 2 described later.

  The data string input to the compression means 1 is, for example, data such as character data, vector data, and image data. The compression means 1 searches the input data for the longest match among the partial strings of encoded data. An adaptive dictionary encoding method for encoding, wherein code data excluding the identification flag is output as 1-byte data, and eight identification flags (1 bit) are output collectively as 1-byte data. As a result, the output is in byte units.

  FIG. 2 is a diagram for specifically explaining the above configuration. First, as shown in FIG. 2A, the input data buffer 3 is composed of a P buffer 4 and a Q buffer 5, and the P buffer 4 is a history buffer and is composed of 32 bytes. An input character string is supplied to the Q buffer 5 and is composed of 9 bytes. A thick frame A shown in the figure is a sliding window, and is configured to be slidable in the direction of arrow c shown in the figure.

  As described above, the input character string is data such as character data, vector data, and image data. Specifically, the input character string is input to the Q buffer 5 by sliding the sliding window A in the direction of arrow c. I do.

  FIG. 2B is a diagram showing a specific configuration of the compressed data. The first byte is flag data, and eight compressed data (and uncompressed data) are recorded after the flag data. The flag data is 1-byte 8-bit data as shown in FIG. 5C, where data “0” indicates uncompressed data and data “1” indicates compressed data. Also, # 1 to # 8 shown in FIG. 6C correspond to # 1 to # 8 attached to the compressed data (and uncompressed data) following the flag data.

  For example, when the first flag data (# 1) is the flag “1”, the first (# 1) data of the 8 compressed data and the uncompressed data is 1-byte (8-bit) compressed data. . The data structure shown in FIG. 5F shows an example of the compressed data, in which the upper 5 bits record the data at the match start position, and the lower 3 bits record the length data of the matched character string. Therefore, the P buffer 4 is a buffer having a capacity that can be expressed by 5 bits, and can perform a matching search of data of a 9-byte character string.

  On the other hand, when the first flag data (# 1) is the flag “0”, the first (# 1) data of the eight compressed data (and uncompressed data) is 1 byte (8 bits) uncompressed. It is data. The data structure shown in FIG. 5 (e) shows an example of the uncompressed data, and is described as the original data.

  The same applies to the compressed data (and uncompressed data) after # 2, and compressed data or non-compressed data corresponding to the state of the flag recorded in the flag data is recorded. The flag data following the above eight compressed data (and uncompressed data) is recorded with information of the following compressed data (and uncompressed data) # 9 to # 16 as shown in FIG. Has been.

  In this example, the P buffer (history buffer) and Q buffer of the slide dictionary are made small, and the encoded data (excluding the identification flag) of the character string reference is output as byte data as 1 byte data.

  Therefore, according to this example, by reducing the slide dictionary of the first compression means, the compression effect is slightly reduced, but the processing speed can be increased. Also in the hardware configuration, the “slide dictionary P-buffer” of the history array and the shift register, the comparators, and the like can be greatly reduced, and the circuit scale can be reduced.

In addition, since the second compression means is provided, the amount of compression efficiency can be reduced by the second compression process, and a high-compression and high-efficiency compression apparatus can be realized.
In addition, by handling all the first compression means with byte data, both the first and second compression means can be handled in units of bytes, bit operations can be reduced as much as possible, and high speed can be achieved.
(Embodiment 2)
Next, Embodiment 2 of the present invention will be described.

  FIG. 3 is a system configuration diagram illustrating the compression device according to the present embodiment. As in FIG. 1, the compression means 11 is a first compression means using an adaptive dictionary type encoding method (Lempel-Ziv method), and the compression means 12 is a semi-adaptive type. ) Second compression means using a probability statistical type encoding method. However, the compression means 11 of this example is configured to simultaneously perform the appearance frequency counting process used by the compression means 12. This will be specifically described below.

  Also in this example, as in the above-described embodiment, input data is input to the compression unit 1 and subjected to encoding processing. The data string input to the compression unit 1 is, for example, character data, and the compression unit 1 searches for and encodes the longest matching partial sequence of encoded data in the dictionary. The encoded data is supplied to an intermediate data buffer 13 composed of a RAM, and when a predetermined amount of encoded data is input, it is output to the compression means 12.

  On the other hand, the compression unit 11 performs a count process of the appearance frequency necessary for the compression process performed by the compression unit 12 and outputs the count result to the frequency table 14. That is, a frequency table used by the compression unit 12 is generated, and a compression process using the probability statistical coding method performed by the compression unit 12 is configured to be performed efficiently.

As described above, according to this example, when the code data is generated by the compression unit 11 as the first compression unit, the appearance frequency is counted at the same time, so that the compression unit 12 as the second compression unit can perform the frequency table. The data can be used for compression processing, and the compression processing can be speeded up.
(Embodiment 3)
Next, a third embodiment of the present invention will be described.

  FIG. 4 is a diagram for explaining the compression processing of this embodiment. In the figure, an input data string such as the aforementioned character data is supplied to the input buffer 20, and this input data is further supplied from the input buffer 20 to the compression means 21 which is the first compression means. The compression means 21 is a compression means using an adaptive dictionary type encoding method as described above, and also performs appearance frequency counting processing.

  The compression means 21 encodes the input data, outputs it to the intermediate data buffer 23, counts the size of the output data of the first data compression means, and appears when generating the encoded data as described above. Count the frequency at the same time.

  The frequency table 24 is a table for recording the appearance frequency, counts up the count data output from the compression means 21, and records the data of the appearance frequency for each character code, for example. The size calculation / determination unit 25 calculates a compressed data size (compression ratio) when the compression unit 22 as the second compression unit performs the compression process, and the output data amount is larger than the size after the first compression. Determine if it increases.

  When it is determined that the amount of output data increases, magic number 1 is output as identification data. On the other hand, when it is determined that the amount of output data does not increase, magic number 2 is output as identification data.

  When the size calculation / determination unit 25 determines that the amount of output data increases, the first compression recorded in the intermediate data buffer 23 with respect to the original data output unit 26 without performing the second compression process. Instructs the output data buffer 27 to output the result. On the other hand, when it is determined that the amount of output data does not increase, the data string of the first compression result recorded in the intermediate data buffer 23 is input to the compression unit 22 as the second compression unit, and the second compression is performed. Instruct to process.

  FIG. 5 is a diagram showing an example of the identification data. Two magic numbers are prepared. The magic number 1 shown in FIG. 5A is compressed by the first compression method, and the second compression method is used. The identification data when processing is not performed is shown. On the other hand, magic number 2 shown in FIG. 5B indicates that the compression processing is performed by the first compression method and the compression processing is further performed by the second compression method. The identification data may be composed of a magic number indicating that compression is performed by the present compression method and a flag indicating whether or not the second compression is performed, instead of the two magic numbers.

Through the above processing, an output corresponding to the identification data is supplied to the output data buffer 27, and the compressed data is output from the output data buffer 27.
As described above, according to this example, after performing the first compression process, the compressed data size (compression ratio) when the second compression process is performed is calculated, and the output data after the first compression process is calculated. When the data amount increases from the size, it is possible to avoid performing the compression process in which the size increases from the result of the first compression process without performing the second compression. Further, it is not necessary to perform the useless second compression process, and an efficient compression process is possible.
(Embodiment 4)
Next, a fourth embodiment of the present invention will be described.

  FIG. 6 is a diagram for explaining the compression processing of this embodiment. In this example, another appearance frequency table is provided with respect to the above-described third embodiment to enable more efficient compression processing. This will be specifically described below. The same components as those in FIG. 4 will be described using the same numbers.

  As described above, an input data string such as character data is supplied to the input data buffer 20, and the input data supplied to the input data buffer 20 is further supplied to the compression means 21 which is the first compression means. The compression unit 21 is a compression unit using an adaptive dictionary type encoding method, performs compression processing of input data, and outputs encoded data to the intermediate data buffer 23. Further, as described above, the count of each size of the input data string and the encoded data string after the first compression and the appearance frequency of the data of each data string are simultaneously performed.

  In this example, the frequency table has two frequency tables 24a and 24b, and the frequency tables 24a and 24b record appearance frequency data for each character code, for example, as described above. The size calculation / determination unit 25 calculates a compressed data size (compression ratio) when the compression unit 22 as the second compression unit performs the compression process, and outputs the output data from the size after the first compression. Determine if the amount increases.

  The size calculation / determination means 25 knows the sizes (1) to (4) below and performs the following processing. That is, (1) the size of the input (original) data string, (2) the output size of the first compression means, (3) the output size when the input (original) data string is compressed only by the second compression means, (4) Output size when the output data string of the first compression means is further compressed by the second compression means.

  First, when (1) is minimized, the input data is output as it is. That is, the size calculation / determination means 25 outputs magic number 1 as identification data, and instructs the original data output means 26 to output the data string input to the input data buffer 20. Therefore, the input data supplied to the input buffer 20 is output to the original data output means 26 without performing the first compression process and the second compression process, and further from the original data output means 26 via the output data buffer 27. Output.

  Next, when (2) is minimized, the first compression process is performed, and the data string is output without performing the second compression process. That is, in this case, the size calculation / determination means 25 outputs the magic number 2 as the identification data and instructs the original data output means 26 to output the data string recorded in the intermediate data buffer 23. Therefore, in this case, the encoded data stored in the intermediate data buffer 23 is output via the original data output means 26 and the output data buffer 27.

  Next, when the above (3) is minimum, the input data is output after being subjected only to the second compression process. That is, in this case, the size calculation / determination means 25 outputs the magic number 3 as identification data, causes the second compression means 22 to input the data string from the input data buffer 20, and performs the second compression processing. Instruct to apply. Therefore, based on this instruction, the second compression means 22 inputs the data string of the input data buffer 20, performs the second compression process, and outputs the data to the output data buffer 27.

  Next, when (4) is minimized, the first and second compression processes are performed on the input data and output. That is, in this case, the size calculation / determination means 25 outputs the magic number 4 as the identification data, and instructs the second compression means 22 to perform compression processing. Therefore, based on this instruction, the compression means 22 further compresses the encoded data recorded in the intermediate data buffer 23 and outputs the compressed data via the output data buffer 27.

  The meaning of each magic number in the identification data is as shown in FIG. 7A. Four magic numbers are prepared. Magic number 1 is compressed by this compression method and both the first and second compressions are performed. Indicates that the compression is not performed, magic number 2 is compressed by the present compression method and the first compression is performed, second compression is not performed, and magic number 3 is compressed by the present compression method and the first compression. Indicates that the compression has not been performed and the second compression has been performed, and the magic number 4 has the meaning that the compression has been performed by the present compression method and the second compression has been performed.

  As for identification data, as shown in FIG. 4B, instead of the four magic numbers, a magic number indicating that compression has been performed by this compression method and whether or not the first compression has been performed. It is good also as a structure comprised with the flag which shows whether the flag shown and the 2nd compression were performed / not performed.

As described above, according to this example, the two means of the first and second compression can be combined to provide the compression result with the highest compression efficiency, and the worst case prevents the increase from the input data size. it can.
(Embodiment 5)
Next, a fifth embodiment of the present invention will be described.

  FIG. 8 is a diagram for explaining the compression processing of the present embodiment. Unlike the first to fourth embodiments described above, this example describes compression processing by semi-adaptive stochastic statistical coding. The compression means 31 of this example includes a statistical model processing means 32 and an entropy coding means 34. It consists of. This will be specifically described below.

  First, input data such as the above-described character data is supplied to the input data buffer 30 and, for example, compressed data compressed by the above-described first compression unit is supplied. The input data supplied to the input data buffer 30 is supplied to the statistical model processing means 32. The statistical model processing means 32 counts the appearance frequency of each byte data. The count result of the appearance frequency is supplied to the frequency table 33, and the count result of the appearance frequency is stored in the frequency table 33.

  The entropy encoding unit 34 includes a preparation processing unit 35, an entropy encoding processing unit 36, an output data buffer 1 (F1 table) to an output data buffer 8 (F8 table), and a code data output processing unit 37.

  The preparation processing unit 35 calculates the compressed data size and outputs the allocation process of the output data buffer 1 (F1 table) to the output data buffer 8 (F8 table) and the codes of the best modes 1 to 23 described later to the header unit. Perform the process. The entropy encoding processing unit 36 generates a compression format based on the best mode information created by the preparation processing unit 35.

  First, the processing of the preparation processing unit 35 will be described. FIG. 9 is a diagram for explaining the configuration and processing outline of the preparation processing unit 35, and a in FIG. 9 is the frequency table 33 described above. The frequency table 33 stores a counter value (frequency) corresponding to the character code (byte data (00 to FF)), and the preparation processing unit 35 uses the counter value (frequency) to determine the best (hereinafter, A storage area b for storing a code having a higher frequency than 23 (indicated by “Best”) and the frequency (counter) of the code is created.

  FIG. 10 is a flowchart for explaining processing performed by the preparation processing unit 35. First, the code table and frequency table stored in the storage areas a and b are initialized, and the total frequency number is cleared to 0 (step (hereinafter referred to as ST) 1). Next, it is determined whether the data in the frequency table is 0 (ST2), it is confirmed that the data exists in the frequency table 33 (ST2 is NO), and the data is extracted from the frequency table 33 (ST3).

  Next, the extracted frequency data is accumulated in the total frequency number (ST4). Then, it is determined whether the code frequency of Best 23 is 0 (ST5). Here, when the code frequency of Best23 is not 0 (ST5 is NO), it is compared with the code frequency of Best23 (ST6), and when it is less than the frequency of Best23, it is determined that the character code does not appear frequently, Return to the decision (ST2). On the other hand, if the frequency is more than the frequency of Best 23, the process proceeds to processing (ST7).

  If the frequency matches the code of Best 23, it is determined whether it is larger than the code of Best 23 (ST8). That is, even if the appearance frequency is the same, since the code number of the character code is a young code, if it is larger than the code number of Best 23, the process returns to the determination (ST2). On the other hand, if it is smaller (younger) than the code number of Best 23, the process proceeds to processing (ST7).

  In the process (ST7), the extracted frequency and code are inserted into the frequency table and code table of Best 1 to 23 in ascending order of frequency. By this processing, the data stored in the storage area of FIG. 9A is rearranged in order of frequency and stored in the frequency counters and codes of Best 1 to 23 shown in FIG.

  Next, when all the data of the frequency table is read and the end is reached (YES in ST2), the codes of Best 8 to 23 are further rearranged in ascending order (ST9). Further, the size calculation of the output data buffer 1 (F1 table) to the output data buffer 8 (F8 table) is performed (ST10).

For example, the size of each table is calculated based on the following calculation. That is,
F1 table size = (number of input data + 7) / 8 bytes
F2 table size = (sum of frequencies from 1 to 23 + 7) / 8 bytes
F3 table size = (sum of frequencies from 2 to 23 + 7) / 8 bytes
F4 table size = (sum of frequencies from 4 to 23 + 7) / 8 bytes
F5 table size = (sum of frequencies from 2 to 3 + 7) / 8 bytes
F6 table size = (sum of frequencies from 4 to 7 + 3) / 4 bytes
F7 table size = (sum of frequencies of Best 8 to 23 + 1) / 2 bytes
F8 table size = number of input data−sum of frequencies of Best 1 to 23 Next, output data buffer 1 (F1 table) to output data buffer 8 (F8 table) are assigned to compression formats from the above calculation results (ST11). Also, data including code information of Best 1 to 23 is output to the header portion of the compressed format (ST12, ST13).

  FIG. 11 is an example of a compression format. This compression format is composed of a header part and F1 to F8 tables, and the header part is composed of a total compression capacity, the number of bits of the F1 to F7 tables, the number of bytes of raw data, and the code tables of Best1 to 23.

  Based on the data supplied from the preparation processing unit 35, the entropy encoding processing unit 36 sets specific numerical values for the compression format according to the flowchart shown in FIG.

  First, 1 byte of the input data supplied to the input buffer 30 is read (step (hereinafter referred to as STP) 1), and it is determined whether it is a data end (STP2). If input data exists (STP2 is YES), information on the appearance frequency of the read data is extracted (STP3).

  Next, it is determined whether the read 1-byte data is a Best1 code (STP4). Here, when the read 1-byte data is the Best1 code (STP4 is YES), the Best1 code is processed (STP5). That is, the bit of the output data buffer 1 (F1 table) is turned on, and the bit of the output data buffer 2 (F2 table) is turned off.

  When the read 1-byte data is not the Best1 code (STP4 is NO), it is determined whether or not it matches the code of Best2 (STP6). Here, when the code matches the code of Best2, the code process of Best2 is performed and the bit of the output data buffer 5 (F5 table) is turned off (STP7). Further, the bit of the output data buffer 1 (F1 table) is turned on, the bit of the output data buffer 2 (F2 table) is turned on, and the bit of the output data buffer 3 (F3 table) is turned off (STP8).

  Hereinafter, according to the flowchart shown in the figure, it is determined whether or not the codes of the rankings Best3, Best4, Best5, Best6, and Best7 match, and if they are the same code, the corresponding processing (STP9 to STP19) is executed.

  Next, the code frequency of Best23 is compared with the frequency of the read code (STP20), and if it is less than the frequency of Best23, it is determined that the character code does not appear frequently, and uncompressed data processing is performed (STP21). . That is, the bit of the output data buffer 1 (F1 table) is turned off, and the read code data is set in the output data buffer 8 (F8 table).

  On the other hand, if the frequency of the code of Best 23 is greater, the code table of Best 8 to 22 is searched for 2 minutes, and the table No. Is set in the output data buffer 7 (F7 table) (STP22).

  If the frequency of the code of Best 23 matches, the code of Best 23 is compared with the code (STP 23). If they match, the code number of Best 23 is set in the output data buffer 7 (F7 table) (STP 24).

  Next, code processing of Best 8 to 23 is performed (STP25). That is, the bit of the output data buffer 1 (F1 table) is turned on, the bit of the output data buffer 2 (F2 table) is turned on, the bit of the output data buffer 3 (F3 table) is turned on, and the output data buffer 4 (F4) Turn on the table bit.

  The above processing is repeated until the input data reaches the end, and finally the code data output processing means 37 sequentially outputs the output data buffer 1 (F1 table) to the output data buffer 8 (F8 table).

By processing as described above, the output format is variable length encoding of 2 bits for the Best 1 code, 4 bits for the Best 2 to 3 code, 6 bits for the Best 4 to 7 code, 8 bits for the Best 8 to 23 code, and 9 bits for the uncompressed data, and The output data area is divided into eight areas of output data buffer 1 (F1 table) to output data buffer 8 (F8 table), and memory access for encoding and decoding is made in 1-bit units while being a variable length code. It is possible to perform bit access in units of 2 bits and 4 bits and byte access of 1-byte data, and very high-speed compression processing can be performed.
(Embodiment 6)
Next, a sixth embodiment of the present invention will be described.

  FIG. 13 is a diagram for explaining the compression processing of the present embodiment. In addition, although this example demonstrated the compression process by stochastic statistical type encoding in the said Embodiment 5, in this example, the compression efficiency which calculates | requires the compression size for every output format by calculation, and determines the case where compression efficiency is the best The point provided with the determination part is the feature. This will be specifically described below.

  FIG. 13 is a diagram for explaining this example, and the same components as those in FIG. First, input data such as the aforementioned character data is supplied to the input data buffer 30, and the input data supplied to the input data buffer 30 is supplied to the statistical model processing means 32. The statistical model processing means 32 counts the appearance frequency of each byte data. The count result of the appearance frequency is supplied to the frequency table 33, and the count result of the appearance frequency is stored in the frequency table 33.

  The entropy encoding unit 34 includes a preparation processing unit 35, an entropy encoding processing unit 36, a compression efficiency determination unit 39, an output data buffer 1 (F1 table) to an output data buffer 8 (F8 table), and a code data output processing unit 37. It consists of

  The preparation processing unit 35 calculates the compressed data size, and outputs the allocation process of the output data buffer 1 (F1 table) to the output data buffer 8 (F8 table), the codes of the best modes 1 to 23 described later, and the like to the header part. Process. Further, the entropy encoding processing unit 36 performs a process to be described later for each data on the input data based on the best mode information created by the preparation processing unit 35.

  The compression efficiency determination unit 39 calculates the compressed data size for each compression format based on the data supplied from the frequency table 33, and determines the output format having the minimum size. The preparation processing performed by the preparation processing unit 35 is executed according to the flowchart shown in FIG. Further, the compression efficiency determination unit 39 performs the same processing as the processing performed by the entropy encoding processing unit 36, calculates the compressed data size for each compression format, and determines the output format having the minimum size.

FIG. 14 is a diagram illustrating an example of a determination pattern.
(1) In the case of the Best1 code and other codes, Best1 is encoded by 1 bit, and the others are encoded by 9 bits.
(2) In the case of Best1, next Best2, and other codes, Best1 is encoded by 2 bits, Best2 is encoded by 3 bits, and others are encoded by 9 bits.
(3) In the case of Best1, next Best4, and other codes, Best1 is encoded with 2 bits, Best4 is encoded with 4 bits, and others are encoded with 9 bits.
(4) For Best1, next Best16, and other codes, Best1 is encoded with 2 bits, Best16 is encoded with 6 bits, and others are encoded with 9 bits.
(5) For Best1, next Best2, next Best4, and other codes, Best1 is 2 bits, Best2 is 4 bits, Best4 is 5 bits, and others are 9 bits.

(6) In the case of Best1, next Best2, next Best16, and other codes, Best1 is 2 bits, Best2 is 4 bits, Best16 is 7 bits, and the others are 9 bits.
(7) In the case of Best1, next Best4, next Best16 and other codes, Best1 is 2 bits, Best4 is 5 bits, Best16 is 7 bits, and others are 9 bits.
(8) For Best1, next Best2, next Best4, next Best16 and other codes, Best1 is 2 bits, Best2 is 4 bits, Best4 is 6 bits, Best16 is 8 bits, others are 9 bits Is done.

  The specific bit contents are the data and code shown in the figure, respectively. For example, (1) Best1 code and otherwise, byte data matching the Best1 code is stored in the output data buffer 1 (F1 table). It is stored as a flag indicating the Best1 code in a 1-bit storage area, and is encoded with 1 bit. Data other than the Best1 code is stored as a flag indicating uncompressed data in the 1-bit storage area of the output data buffer 1 (F1 table), and is stored in the raw data storage area of the output data buffer 8 (F8 table). Data (8 bits) is stored and encoded with a total of 9 bits. Hereinafter, (2) Best1, next Best2, and other codes are also shown in the figure.

  FIG. 15 shows the output format of the compression format that the compression efficiency determination unit 39 determines and outputs from the calculation result. From the calculation result, for example, the output format of “(1) Best1 code and other codes” is determined, and when selected, the compression format with the smallest size can be used. For example, in the case of image information of the same color, it is possible to select a compression format that is extremely suited to the characteristics of the input data.

Therefore, according to this example, it is possible to select the output format having the best compression efficiency corresponding to the input data and perform the compression processing, and it is possible to shorten the compression processing time and the transfer time of the compressed data, which is extremely efficient. Output compression processing can be performed.
(Embodiment 7)
Next, a seventh embodiment of the present invention will be described. In this example, the output format 4 determined by the compression efficiency determination unit 39 (that is, (4) Best1, next Best16, and other cases) is used. This is an example in the case where frequency data from Best 1 to Best 17 is created by the preparation processing unit 35, and the compression format has the configuration shown in FIG. Therefore, the table numbers in this example do not match the output data buffer (F table) in the sixth embodiment.

  FIG. 17 is a flowchart showing processing performed by the entropy encoding processing unit 36 in this example. First, one byte is read from the input data supplied to the input data buffer 30 (step (hereinafter referred to as W) 1), and it is determined whether it is a data end (W2). In this first process, there is sufficient input data (W2 is YES), and information on the appearance frequency of the read data is extracted (W3). That is, the output data buffer 1 (F1 table) of this example is searched to determine whether the read 1-byte data is the Best1 code (W4).

  Here, when the read 1-byte data is the Best1 code (W4 is YES), the Best1 code processing is performed (W5). That is, the bit of the output data buffer 1 (F1 table) of this example is turned on, and the bit of the output data buffer 2 (F2 table) is turned off.

  Further, when the read 1-byte data is not the Best1 code (W4 is NO), a comparison process with the code of Best17 is performed (W6). If the code frequency of Best 17 is less, uncompressed data processing is performed, the bit of the output data buffer 1 (F1 table) of this example is turned off, and the read data is set in the output data buffer 4 (F4 table). (W7). On the other hand, if the frequency of the code of Best 17 is higher, the code table of Best 12 to 16 is searched for 2 minutes, and the table number of the matched code is set in the output data buffer 3 (F3 table) (W8).

  If the frequency of the code of Best17 matches, the code of Best17 is compared with the code (W9), and if it matches, the value of the code of Best17 is set in the output data buffer 3 (F3 table) (W10). And the code processing of Best12-17 is performed. That is, the output data buffer 1 (F1 table) is turned on, and the output data buffer 2 (F2 table) is turned on.

As described above, according to this example, it is possible to select the output format having the best compression efficiency corresponding to the input data and perform the compression processing, and it is possible to shorten the compression processing time and the compressed data transfer time. . In this example, the appearance frequency of the character code has been described in the examples from Best 1 to 17. However, the appearance frequency may be from Best 1 to 23 as shown in the above-described sixth embodiment, and will be described in other examples. May be.
(Embodiment 8)
Next, an eighth embodiment of the present invention will be described.

  FIG. 18 is a diagram for explaining the present embodiment. Before searching for the longest matching character string, it is determined whether or not the same character as the immediately preceding character is continuous. If it is, the process of compressing the run length character is executed. For example, FIG. 5A shows a case where eight consecutive characters “a” are consecutive, and in this case, compressed data of flag + length is used. Further, the flag is also used as the longest matching character string reference, and as shown in FIG. 19 (g), identification information for continuous length compression is entered in the "matching start position" portion of the character string reference code code. Note that FIG. 18B is the same as FIG. 2A described above, and a description thereof will be omitted. Also, the configurations of FIGS. 19B to 19F are the same as the configurations of FIGS. 2B to 2F described above, and a description thereof will be omitted.

For example, when there are many white parts in the image data, these parts are continuous with the same value. Therefore, compression efficiency can be improved by adding the continuous length character compression of this example to such input data.
(Embodiment 9)
Next, a ninth embodiment of the present invention will be described.

  FIG. 20 is a diagram illustrating this embodiment. This example includes original data division / input means 40 for dividing and controlling an input data string as a response to a large amount of input data, and the input data stored in the input file 41 is divided and input by the original data division / input means 40. It is configured to output to the data buffer 43.

With this configuration, it is possible to easily cope with a large amount of input data by dividing the data and supplying it to the input data buffer 43.
The other configuration shown in the figure is the same as the configuration shown in FIG. The compressed data stored in the output data buffer 27 is sequentially output to an output file by the compression result data output means 44.
(Embodiment 10)
Next, a tenth embodiment of the present invention will be described.

  In the first to ninth embodiments described above, the compression processing of input data has been described. In the following embodiments, the decompression processing of data compressed by the above-described compression processing will be described. This will be specifically described below.

  In FIG. 21, the first restoration means 51 decodes semi-adaptive probability statistical type encoded data, and the second restoration means 52 uses adaptive dictionary type encoded data (Lempel-Ziv). Method). That is, the compressed data decompression process performed in the first to ninth embodiments is performed.

  Therefore, the compressed data performed in the above-described first to ninth embodiments is input to the first decompression unit 51 as an input data string, and is compressed by the probability statistical encoding method of the above-described second compression unit. Decrypt the column. The second restoration means 52 decodes the data string compressed by the adaptive dictionary type encoding method (Lempel-Ziv method) of the first compression means.

With this configuration, reversible (lossless) decoding is performed, and composite processing corresponding to compression processing can be performed.
(Embodiment 11)
Next, an eleventh embodiment of the present invention will be described.

This embodiment also explains data restoration processing. This will be specifically described below.
FIG. 22 is a diagram for explaining the data restoration apparatus of this example. In the figure, a first restoration means 51 performs a decoding process of semi-adaptive probability statistical type encoded data as described above, and a second restoration means 52 performs adaptive dictionary type encoded data. (Lempel-Ziv method) decryption processing is performed. In the same figure, the input switching means 53 inputs the data string encoded (compressed) by the processing of the third embodiment, and the identification data determination means 54, the first decoding means 51, or This is input data switching means for transferring data to one of the second decoding means 52. In the initial state, the input switching means 53 is set to pass input data to the identification data determination means 54.

  Here, the identification data used in this example is the magic number described in the third embodiment, and the identification data determination unit 54 determines the received input data as the identification data. That is, the following determination is made in comparison with the magic number 1 or 2 described above.

  For example, if the magic number is 1, the input switching means 53 is switched and the input data is supplied to the second decoding means 52 via the switching means 55. At this time, the second decoding means 52 is instructed to start the decoding process.

  If the magic number is 2, the input switching means 53 is switched, the input data is supplied to the first decoding means 51, the switching means 55 is further switched, and the output data of the first decoding means 51 is the second decoding means. Control is performed so as to be supplied to 52. Then, the first decoding means 51 is instructed to start the decoding process. The second decoding unit 52 is also instructed to start the decoding process.

Therefore, according to this example, the decoding process can be performed based on the magic number set at the time of compression, and the decoding process of the compressed data can be efficiently performed using the decoding method corresponding to the compression method.
Embodiment 12
Next, a twelfth embodiment of the present invention will be described.

This embodiment also explains data restoration processing. This will be specifically described below.
FIG. 23 is a diagram for explaining the data restoration apparatus of this example. In the figure, a first restoration means 51 performs a decoding process of semi-adaptive probability statistical type encoded data as described above, and a second restoration means 52 performs adaptive dictionary type encoded data. (Lempel-Ziv method) decryption processing is performed. In the same figure, the input switching means 53 and the identification data determination means 54 have the same configuration as described above, but the identification data determination means 54 identifies four types of magic numbers described later. That is, this example is a configuration that restores the compressed data by the compression device of the above-described fourth embodiment, and the identification data performs the determination of the above-described four types of magic numbers 1 to 4.

In this example, two switching means 55a and 55b are used, and an output switching means 58 is also provided.
In the above configuration, when the identification data to be input to the identification data determination unit 54 is the magic number 1, the identification data determination unit 54 outputs the data string passed as it is to the data end, excluding the determined identification data. To do.

  If the magic number is 2, the identification data determination unit 54 instructs the input switching unit 53 to supply the input data to the second decoding unit 52, and switches the input switching unit 53 and the switching unit 55a. Further, it instructs the output switching means 58 to output the data string from the second decoding means 52. Therefore, by this process, the input data is decoded by the second decoding means 52 and the decoded data is output via the output switching means 58.

  If the magic number is 3, the identification data determination unit 54 switches the input switching unit 53 and instructs the first decoding unit 51 to supply the input data. Further, the switching means 55b and the output switching means 58 are switched to instruct the first decoding means 51 to start decoding. Therefore, by controlling in this way, the first decoding unit 51 starts the decoding process, and the decoded data is output via the switching unit 55b and the output switching unit 58.

  Further, if the magic number is 4, the identification data determination unit 54 is further configured to use the input switching unit 53, the switching unit 55a, and the second decoding unit 52 to decode the data string decoded by the first decoding unit 51. 55b and the output switching means 58 are switched. Then, the first decoding unit 51 and the second decoding unit 52 are instructed to start decoding. Therefore, based on this instruction, the first decoding means 51 and the second decoding means 52 start the decoding process, and the decoded data decoded by both the decoding means 51 and 52 is output via the output switching means 58. The

  If the magic numbers are not 1 to 4, the identification data determination unit 53 outputs the input data string as it is, including the determined identification data. However, in this case, an error response may be adopted.

By performing the processing as described above, decoding processing can be performed based on the magic numbers 1 to 4 set at the time of compression, and efficient decoding processing is performed by selecting an appropriate decoding method corresponding to the compression method. Can do.
(Embodiment 13)
Next, a thirteenth embodiment of the present invention will be described.

  This embodiment also explains data restoration processing. In this example, the data string compressed by the probability statistical encoding method according to the fifth to seventh embodiments is restored. This will be specifically described below.

  FIG. 24 is a flowchart showing the decoding algorithm of this example. The encoded data string is output by dividing the F1 to F8 table, flag, and data included in the above-described compression format into eight areas (or four areas).

  First, preparation is made to extract data from the F1 to F8 tables (step (hereinafter referred to as V) 1). Next, the data end of the F1 table is determined (V2). If data exists, 1-bit data is extracted from the F1 table (V3), and it is determined whether the data is uncompressed data or compressed data (V4). ).

  If the data is uncompressed data, 1-byte data is extracted from the F8 table and output (V5). On the other hand, if the data is compressed data (V4 is YES), 1-bit data is extracted from the F2 table, and it is determined whether the data has the highest appearance frequency (V7). If the data has the highest appearance frequency, the Best1 code is output with reference to the Best1 code (V8).

  If the code is not the Best1 code, the F3 table is referred to in units of bits, and it is determined whether the code is the Best2-3 code having the next highest appearance frequency (V10). Here, when it is a code of Best 2 to 3 having the next highest appearance frequency (V10 is YES), the F5 table is referred to bit by bit to determine whether it is a code of Best 2 or a code of Best 3 (V12). If it is the code of Best2, the code table is referenced and the code of Best2 is output (V12 is YES, V13). If it is the code of Best3, the code table is referenced and the code of Best3 is output (V12 is NO) V14).

  On the other hand, in the determination of the F3 table, when the appearance frequency is not the code of Best 2 to 3, the F4 table is referred to in bit units, and the code of Best 4 to 7 with the next highest appearance frequency is determined (V15, V16). . When the code is Best 4-7, the F6 table is accessed in units of 2 bits, and the corresponding data in the code table of Best 4-7 indicated by the 2 bits is output (V16 is YES, V17). On the other hand, if it is not the code of Best 4-7, the F7 table is accessed in units of 4 bits, and the corresponding data in the code table of Best 8-23 indicated by the 4 bits is output (V16 is NO, V18).

By performing the data restoration process as described above, the data string compressed by the probability statistical coding method can be restored, and the data restoration process corresponding to the compression process can be performed.
(Embodiment 14)
Next, a fourteenth embodiment of the present invention will be described.

This embodiment also explains data restoration processing. This example corresponds to the ninth embodiment described above.
FIG. 25 is a diagram for explaining this embodiment. In this example, a process for restoring compressed data obtained by dividing and compressing a large amount of input data will be described. In the figure, the original data dividing / inputting means 60 divides the input data stored in the input file 61 and outputs it to the input data buffer 63. With this configuration, the compressed data is divided and supplied to the input data buffer 63, and thereafter the identification data is identified by the identification data determination unit 65, and the first decoding unit 66, the second decoding unit 67, or the original data is identified. The data output means 68 can be selected to decode the compressed data.

  Further, the data decoded by the decoding process of this example is output via the output data buffer 69. During this output process, the decoded data is sequentially output to the output file 71 by the decoding result data output means 70. Is remembered.

By performing the data restoration process as described above, it is possible to obtain a restoration output in a state where even a huge amount of data is divided into the same length as that at the time of compression.
(Embodiment 15)
Next, a fifteenth embodiment of the present invention is described.

  The present embodiment is a modification of the decoding process of the tenth embodiment, and FIG. 26 shows the configuration of this example. In the figure, a first restoration means 71 performs a decoding process of semi-adaptive probability statistical coding data, and a second restoration means 72 performs adaptive dictionary-type coded data (Lempel-Ziv). Method). Further, in this example, a first-in first-out (FIFO) memory 73 is mounted, and when 1-byte compressed data is decoded by the first decompression means 71, 1-byte decoded data is input to the FIFO memory 73. The 1-byte data is output from the FIFO memory 73 to the second restoring means 72 and decrypted.

  Therefore, according to this example, the first restoration process and the second restoration process can be operated in parallel, and the speed of data restoration (decoding) can be increased.

DESCRIPTION OF SYMBOLS 1, 2 ... Compression means 3 ... Input data buffer 4 ... P buffer 5 ... Q buffer 11, 12 ... Compression means 13 ... Intermediate data buffer 14 ... Frequency table 20. .. Input buffers 21, 22 ... Compression means 23 ... Intermediate data buffers 24, 24a, 24b ... Frequency table 25 ... Size calculation / determination means 26 ... Original data output means 27 ... Output data buffer 30 ... Input buffer 31 ... Compression means 32 ... Static model processing means 33 ... Frequency table 34 ... Entropy encoding means 35 ... Preparation processing unit 36 ... Entropy code Conversion processing unit 37... Output processing unit 39... Compression efficiency determination unit 40... Original data division / input unit 41... Input file 43. Buffer 44: Compression result data output means 51, 52 ... Restoration means 53 ... Input switching means 54 ... Identification data determination means 55, 55a, 55b ... Switching means 56 ... Switching means,
58 ... Output switching means 60 ... Original data division / input means 63 ... Input data buffer 65 ... Identification data determination means 66, 67 ... Decoding means 68 ... Original data output means 69 ..Output data buffer 70: Decoding result data output means 71, 72 ... Restore means 73 ... FIFO memory 74 ... Output file

Claims (4)

In a data compression apparatus of probability statistical coding that performs variable length coding according to the appearance frequency,
The information for extracting the data with the highest appearance frequency is recorded in the first table and the second table with 2 bits, and the information for extracting the next two pieces of data is recorded with the first to third tables and the first table. Record 4 bits in the table 5 and extract the next 4 more data in 6 bits in the 1st to 4th tables and the 6th table, and the next 16 more Information for extracting data is recorded in the first to fourth tables and the seventh table in 8 bits, and other data is recorded in the first and eighth tables in 9 bits,
Means for recording a data code corresponding to the rank of the appearance frequency in a code table;
A data compression apparatus comprising: When only the most frequently appearing data and other data are included, information indicating whether the data is compressed data is recorded in the first table, and the code of the most frequently occurring data is recorded in the code table. 2. The data compression apparatus according to claim 1, wherein the data compression apparatus is provided. In a restoration apparatus for probability statistical type encoded data that is variable length encoded according to the appearance frequency,
Refers to the first table to obtain information on whether or not the data is compressed data. If the data is not compressed data, the code data recorded in the eighth table is read. If the data is compressed data, the second table is referred to. Information on whether or not the data has the highest appearance frequency, and if the data has the highest appearance frequency, means for reading out the code data of the corresponding appearance frequency from the code table;
If it is not the data with the highest frequency of appearance, further refer to the third table to obtain information on whether or not the two data with the next highest frequency of appearance, If it is data, further refer to the fifth table, obtain information on whether the data is the second most frequently occurring data or the third most frequently occurring data, and code data of the corresponding appearing frequency from the code table Means for reading
If the two occurrence frequencies are not the next most frequent data, further refer to the fourth table to obtain information on whether or not the four occurrence data is the next most frequent occurrence frequency. If there are a large number of four data, further refer to the sixth table to obtain information on whether the appearance frequency is the fourth to seventh most frequent data, and obtain the code data of the corresponding appearance frequency from the code table. Means for reading;
If the data is not the next most frequently occurring 4 data, further refer to the seventh table, obtain information of 16 data having the next most frequently occurring frequency, and code corresponding to the appearing frequency from the code table Means for reading data;
A data restoration apparatus comprising: When only the most frequently appearing data and other data are present, the first table is referred to obtain information on whether or not the data is compressed data, and if the data is not compressed data, the corresponding code data is obtained from the code table. 4. The data restoration device according to claim 3, wherein the data restoration device is read out.