JPH04149766A - Data compressing and restoring system - Google Patents

Abstract

PURPOSE:To simplify codes so as to obtain a high compression rate by making it possible to register the initial values of plural dictionaries corresponding to a history state and adopting also the history of precedently appeared character strings for a character string in coding. CONSTITUTION:A dictionary 10 is constituted to a dictionary group consisting of the prescribed number of dictionaries 10-1 to 10-N less than the total number of character sorts to be processed and all the character sorts are initially registered in each of respective dictionaries 10-1 to 10-N by allocating a reference number to each character. At the time of coding an input character string, a specific dictionary 10-i in the dictionary group is specified and coded in accordance with subordination with a precedently coded character string, i.e. index information indicating the history, and when the inputted character string is not registered, a character string obtained by adding the succeeding one character to the reference number of the coded character string is registered as a new reference number. Thus, the initial registration of plural dictionaries can be simplified by collecting the subordination at the time of coding/decoding and the efficiency of coding can be improved.

Description

[Detailed Description of the Invention] [Summary] Specifies one of multiple dictionaries with an index based on the subordination relationship (history) with the last character of the immediately preceding encoded character string, and inputs the input character string into the specified dictionary. LZW by specifying the reference number of the subsequence that matches the maximum length among the registered subsequences that have already been encoded.
Regarding the data compression and restoration method that encodes character strings into codes and restores character strings from LZW codes, the purpose is to simplify the initial registration of multiple dictionaries and improve encoding efficiency. It consists of a dictionary group consisting of a small predetermined number of dictionaries, and each dictionary contains all character types in one character w.
The configuration is configured so that the initial registration is performed by attaching a reference number to the position.

[Field of Industrial Application] The present invention relates to a data compression and decompression method using an LZW code, which is known as an improved incremental decomposition type of universal code.

In recent years, computers have come to handle various types of data such as character codes, vector information, and images.
The amount of data handled is also rapidly increasing. When handling large amounts of data, by compressing the amount of data by eliminating redundant parts, you can reduce storage capacity and speed up transmission.

Universal encoding has been proposed as a method that can compress various types of data using one method.

Here, the field of the present invention is not limited to character code compression.
Although it can be applied to a variety of data, in the following we will follow the nomenclature used in information theory, and refer to a single word unit of data as a character, and a string of arbitrary words of data. .

A representative method of universal code is the Giblen-Bell (2iv-Lempel) code.
Information Processing, Vol. 26. No. I, 1985
(see year).

2iv-Lempel code has two types: ■Universal type and ■Incremental parsin type.
g) Two algorithms have been proposed.

Furthermore, as an improvement of the universal algorithm, LZ
There are SS codes (T, C, Be l I, “Be
tter 0PM/L Te[Compression
IEEE Trans, on Co5nu
n, , Vol, C0M-34, No, 12.
(See Dec. 1986).

Moreover, as an improvement of the incremental decomposition type algorithm, L Z
There is a W (Lcmpcl-2iy-Welch) code (T, A, Wech, =A Technique
for High-Perlotmxnce Dxt
xCompression, Computer,
(Refer to Jane 1984) 0 Among these codes, LZW is preferred due to its high-speed processing and simple algorithm.
Codes are now used for file compression on storage devices.

[Prior Art] A flowchart of a conventional LZW code encoding algorithm is shown in FIG. 5, and a flowchart of a decoding algorithm is shown in FIG.

First, LZW encoding has a rewritable dictionary, divides the input string into different strings, numbers these strings in the order of their appearance, and registers them in the dictionary. In this method, a string is registered in a dictionary and encoded by representing only the reference number of the longest scattered character string.

Figure 7 shows a specific example of LZW encoding, and Figure 9 shows LZW encoding.
A specific example of ZW decoding is shown, and FIG. 8 shows the contents of a dictionary used in encoding and decoding. In addition, No. 7. 8.
In order to simplify the explanation, FIG. 9 takes as an example a case where a character string consisting of a combination of three characters abc is compressed and restored.

First, the LZW decoding process shown in FIG. 5 will be explained as follows.

Step Sl (hereinafter, [Step J is omitted): A character string consisting of one character for each character is registered in advance as an initial value, and then encoding is started. To encode Sl, a dictionary is searched using the input first character K to obtain a reference number ω, and this is set as a word-initial string (p"efix string).

S2. Read the next character K of the input data.

S3: Check whether character input is completed.

S4: Search whether the dictionary contains the character K read in 82 (ωK) to the initial character string ω obtained in Sl.

S5: If the character string (ωK) is in the dictionary in S4, S
5, replace the character string (ωK) with the reference number ω, and repeat S2
The search for the maximum match length is continued until the character string (ωK) cannot be found in the dictionary.

S6: If the character string (ωK) is not in the dictionary in S4,
Proceed to S6, output the reference number ω for the character found in sl as the code word code (ω), add a new reference number to the character string (ωK) and register it in the dictionary, and then input the character string in S2. K is replaced with the reference number ω, the dictionary address n is incremented, and the process returns to s2 to read the next character K.

A concrete explanation will be given below with reference to FIG. 7.8.

First, the input data 1nput in FIG. 7 is read from left to right. When you enter the first letter a, there are no matching strings in the dictionary other than the letter a, so 0UTPUT C0DEl(
The reference number ω) is output as a code word. Then, let the character a be the initial character string ω.

Next, if you input the second character S, the expanded character string ωKab, which is the addition of this input character to the initial character string ω, is not in the dictionary, so the character S is 0IITPIIT C0DE 2
is output as a code word. Then, in the extended string ω = a
Add reference number 4 to b and register it in the dictionary. In actual dictionary registration, the character string 1b is registered as shown on the right side of FIG. Then, the character S becomes the word-initial character string ω.

If you then input the third character a, the expanded character string ω, which is the initial character string ω added to the character a, = ba = 2a is from Kokichi, which is not in the dictionary, so the character OU DING PI DING C0DE 2 After outputting as a code word, =ba is added to the extended string ω as 2a
, and register it in the dictionary with reference number 5. stop,
The letter a becomes a new word-initial string ω.

For the fourth human-powered character string, = ab is already registered in the dictionary as the code word 4 of i b, so the character string ωK is set as a new initial character string ω, and the fifth character C Input , and create =4c=abe in the expanded character string ω. Since =abe is not registered in the dictionary in this extended character string ω, 0UTPUT C0DE of character string ab = 1b
4 as code word LT: Output and extend character string ω=a
b e is registered in the dictionary in the form of 40 as code word 6. This process continues in the same manner.

Next, the decoding process shown in FIG. 6 will be explained. In this decoding,
Similar to encoding, decoding is started after registering a character string consisting of one character for each character in the dictionary as an initial value.

Sl・Read the first code C0DE and decode the reference number ω. Change the current reference number ω to 01. Dω, and the first code corresponds to one of the single-character reference numbers already registered in the dictionary, so the character D ( which matches the manual reference number ω)
K) is found 2 and the character K is output. Note that the output characters are set to FINcha+ for later exception processing.

S26 Read the next code CO[)E.

S3: Check whether there is a new code, that is, whether code input has ended.

S4・Read code CO [from IE? Decode the kiln number () and set it as INω.

At S5:54, it is checked whether the code C0DE inputted is registered in the dictionary (ω≧n).

86 Normally, the manually generated code word has been registered in the dictionary in the previous processing, so the process advances to S6 and the character string D (ω'K) corresponding to the reference number ω is read from the dictionary.

87 The character string K is temporarily stacked, the reference number ω' is set as a new ω, the process returns to S6, and the procedure of S6 is recursively repeated until the reference number ω reaches the character 1.

S8: The characters stuck at 87 are 1. , IL
Pop up and output the O(Laat In Fast Oυ0 format. At the same time, the reference number OtD used last time is
Combine ω and the first character K of the jointly restored character string (OLD
A new reference number n is added to the character string expressed as ω, K) and registered in the dictionary.

This LZW decoding process will be specifically explained with reference to FIG. 9 as follows.

First, the human code is 1, and the characters are a and b.

Regarding C, reference number 1. 2. 3 is registered in the dictionary as shown in Table 1, so by referring to the dictionary, the character string a1. :Replace and output. Similarly, for the next code 2, the character b
1. l[Replace and output. At this time, a new reference number 4 is added to the previously processed code and the current [mouth 1 decoding (7, first character S, l!:) is added to the (]ab, and the code is registered in the dictionary).

The third code 4 searches the dictionary], replaces b with ab, and outputs the character string ab. At the same time, a character string 2a (=ba), which is a combination of the previously processed code 2 and the character a of the character string 1 and 1 of the currently decoded character string, is registered in the dictionary with a new reference number 5 added thereto.

This process is repeated in the same manner.

In the decoding of FIG. 9, there is the following exception process 6. This exception process occurs in the decoding of the input code 8 of No. 6 L1. code 8
is not defined in the dictionary at the time of decoding and cannot be decoded. In this case, a character string 5b is obtained by adding the first - character of the previously decoded character string ba to the previously processed code 5, which is then replaced with 2ab and bab and output. Then, a reference number 8 is added to a character string 5b obtained by adding the characters of the character string just decoded to the previous code word 5 to the output word of the character string, and the result is registered in the dictionary.

This exception handling is performed at 85 and S9 in the decoding process flow in FIG.
Finally, in step 88, the character string is output and the new character string is registered with a reference number in the dictionary.

The encoding/decoding processes shown in FIGS. 4 and 5 are performed without creating the same dictionary.

[Problem to be solved by the invention] As described above, in the conventional LZW code, when an input character string is divided into different character strings and encoded, each character string currently being encoded is different from the previous character string. is encoded as appearing independently.

This method has no problem in encoding memoryless information sources. However, many data such as actual sentences are considered as memory information sources, and conventional LZW codes cannot fully utilize the history of character string occurrences, and even after data compression, the dependence of character string occurrences cannot be determined. It had the drawback of remaining redundant.

To address these drawbacks, the inventors of the present invention reduced the redundancy between character strings by incorporating the dependency relationship between the encoded character string and the last character of the immediately preceding character string, that is, the history, into a dictionary, and compressed the coded character string. We propose a data compression and decompression method that increases the efficiency.

Specifically, as shown in FIG. 10, the dictionary 10 is divided into a plurality of dictionaries 10-1.1.0-2.10-3.10-4 and indexed. A specific dictionary 10-2 is selected using the last character of ab as an index. In each dictionary 10-1 to 10-4, only character strings that follow the index character are registered.

According to this method, whereas conventionally character strings in a dictionary were specified using reference numbers that looked at the whole string, it is possible to specify using reference numbers only for sequences connected to the index, so small reference numbers can be used, and LZW code can be expressed in a short form to improve encoding efficiency.

However, with this method, the following problem arises in how to set the initial value.

In the LZW code, when data is handled in units of bytes, all 256 character types are registered in the dictionary as initial values in order to simplify the code word and express it only with reference numbers. When this initial registration is applied to the method shown in Figure 10, all character types are 256.
Since a dictionary with 256 children is used as an index and initial registration is performed for each dictionary, it is necessary to register 256x256 (64K) pieces in advance. In reality, many of the 64 characters are not used, so if the initial value is registered as is, the reference numbers of the characters that are not used will become larger and the encoding efficiency will decrease.

Therefore, the inefficiency of registering the initial value can be solved by not registering the initial value in the dictionary in advance and registering it in the dictionary when a character string consisting of one character newly appears. However, when this method is adopted, it is not possible to represent all code words with reference numbers; character strings where the code word consists of one character encode raw data, and character strings with two or more characters are represented by reference numbers. There was a problem in that the algorithm was complicated because it needed to be separated into what was to be encoded and what was to be encoded.

The present invention was made in view of these problems, and
When specifying one of multiple dictionaries for encoding and restoration using an index based on the dependency relationship with the last character of the immediately preceding character string that has been encoded, initial registration of multiple dictionaries can be simplified by combining the dependency relationships. An object of the present invention is to provide a data compression and decompression method that improves encoding efficiency.

[Means to solve the problem] FIG. 1 is a diagram explaining the principle of the present invention.

First, the present invention relates to a data compression method that encodes an input character string into an LZW code by specifying the reference number of a substring that matches the maximum length among already encoded substrings registered in the dictionary 10.

In the present invention, as such a data compression method, the dictionary 10 is constituted by a dictionary group consisting of a predetermined number of dictionaries 10-1 to 10-N, which is smaller than the number of all character types to be processed. All character types are initially registered for each character from -1 to 10-N with a reference number attached to each character.

When encoding an input character string, a specific dictionary 10-i in the dictionary group is designated and encoded according to index information indicating a dependency relationship (history) with previously encoded character strings,
At the same time, if there is no input character string in the specified dictionary 10-i, a character string obtained by adding the next character to the reference number of the previous encoded character string is registered with a new reference number. shall be.

Here, when encoding an input character string, a specific dictionary 10-1 in the dictionary group is designated according to index information obtained from a part of the final character code of the character string encoded immediately before. More specifically, a specific dictionary 0-1 in the dictionary group is specified according to index information indicated by the upper bits of the final character code of the character string encoded immediately before.

On the other hand, when encoding an input character string, a specific word 1!10-1 in the dictionary group is selected according to index information obtained by referring to a lookup table using the final character code of the previously encoded character string. may also be specified. Specifically, a specific word 1f10-i in the dictionary group is specified according to index information obtained by referring to a lookup table using the upper bits of the last character code of the character string encoded immediately before.

In addition, the present invention specifies the input character string by the reference number of the maximum length matching substring among the already encoded substrings registered in the dictionary 10, and generates the original character string from the encoded code word. The data restoration method for restoring the ``Dictionary] 0 is a predetermined number of dictionaries 10- that are smaller than the number of all character types to be processed.
It consists of a dictionary group consisting of 1 to 10-N, and each word 110-
All character types are initially registered in units of 1 to 10N with reference numbers attached to each character. When restoring an input code word, a specific dictionary]O-1 in the dictionary group is specified and restored according to index information indicating the dependency relationship with previously restored character strings, and each time the input code word is restored, The present invention is characterized in that a character string obtained by adding the first character of the currently restored character string to the reference number of the restored character string is registered with a new reference number.

Here, the specification of the specific dictionary 10-1 used at the time of restoration is the same as in the case of encoding.

[Operations] According to the data compression and decompression method of the present invention having such a configuration, the following effects can be obtained.

First, the history showing the dependency relationship with the last character of the previous character string is
As it is, there are 256 states, but characters appear unevenly, and some states do not appear among the 256 states. Therefore, the present invention merges and reduces the history of the final character, resulting in a meaningfully small number of states, for example 8 to 16, without reducing the number of dictionaries.

Since the number of history states is small, the number of registrations of 256 characters of all character types as an initial value of 5 in each dictionary is the number of histories, that is, the number of dictionaries x 256, and it is possible to avoid large waste.

As a method of summarizing the history, for example, by taking the upper 4 bits of the last character of the immediately preceding encoded character string, the history can be summarized into 16 states. In order to effectively use a dictionary, it is desirable to use states that appear evenly when organizing the history. 2. It is not necessarily necessary to use the raw data bits in the characters, but depending on the general nature of the data,
Prepare a lookup table (L[JT) that associates the last character of the previous encoded character string with the history state, and
The history state of the immediately preceding character, that is, the dictionary index may be specified.

[Embodiment] FIG. 2 is a block diagram of a smaller embodiment of the present invention.

In FIG. 2, 12 is a CPU as a control means, and a program memory 14 and a data memory 26 are connected to the CPU 12.

The program memory 14 contains control software 16, L
Maximum-growth search software 18 that performs maximum match length search using ZW codes, encoding software 20 that converts input character strings into ZW codes. decoding software 22 that restores the code converted into the LZW code by the encoding software 20 to the original character string;
and dictionary initial value creation software 24 for initially registering all character types to be processed, for example, 256 character types.

On the other hand, the data memory 26 includes a data buffer 28 for storing a character string to be encoded or a code string to be decoded, and a data buffer 28 for storing a character string to be encoded or a code string to be decoded, and a data buffer 28 for storing a character string to be encoded or a code string to be decoded. It is provided with a dictionary 10 which is used from the beginning.

For example, the dictionary 10 has 16 dictionaries 10-1 for all 256 character types, for example, when classification is performed using index information indicating a dependency relationship consisting of the upper 4 bits of the final character code of an encoded character string. Consists of 1 to 10-16. The dictionary index can be specified directly using the upper 4 bits of the final character code of the encoded character method, but in the following explanation, the dictionary index is read by referring to a lookup table (LUT). Let's take as an example the case where you specify .

The outline of data compression and restoration according to the present invention in the embodiment shown in FIG. 3 is as follows.

The CPU 12 starts the dictionary initial value creation software 24 under the control of the control software 16 to perform dictionary initial value creation processing. Specifically, the dictionary initial value creation software 24
All of the character types 256 are registered in each of the 16 dictionaries 10-1 to 10-16 constituting the dictionary by assigning a reference number to each character.

The data buffer 28 of the data memory 26 stores data to be encoded from the outside for multiple characters of a certain length at once.
The encoder software 20 transfers each character one character at a time according to the request. Then, each time the data buffer 28 becomes empty, a plurality of characters are similarly fetched from the outside.

Next, in FIG. 3, the encoding algorithm of the present invention will be explained with reference to the flowchart of FIG. 3. First, in Sl, the following processing is performed.

■Each of N dictionaries Di (
However, j=1. ..., N), all types of character strings consisting of - characters are registered in advance as initial values. In the present invention, the total number of dictionaries N is as small as N216 for all 256 character types.

(2) The total number of reference numbers of each dictionary Di is managed as 01, and at the time of initialization, n1=(character type+1) is set in nl of the number N of dictionaries.

(2) The history from the immediately preceding character string, that is, the upper 4 bits of the last character code of the immediately preceding character string is set as PK, and PK=0 is set in PK as the initial value.

■Input the first character and change it to the reference number (initial character string) ω.

■Set the LUT that corresponds to the history status from 1 to the last character of the immediately preceding character string. However, at first there is no immediately preceding character string, so to indicate the last character of the immediately preceding character string, set 1 = 0, and use the index PK obtained from the LUT with K1 = 0.
Set the LUT so that PK=Q.

When such processing of Sl is completed, encoding is performed according to the steps from $4 to s7. The procedure from 84 to S7 is basically the same as the conventional one shown in FIG.

The difference is that in the conventional LZW encoding, there was only one dictionary, whereas in the present invention, 1 is used for the last character of the encoded character string shown in Sl at the beginning and S6 thereafter. The history state LOT (KL) obtained by referring to the LUT by
= Select a specific dictionary DP (from multiple dictionaries by PK, search for the maximum growth character string by matching it with the character strings registered in the selected dictionary D□, and increase the maximum match length by one character. The difference is that the column ωK is registered in the selected dictionary I) pg.

After registering in the dictionary D□ in S8, the counter nPK for managing the reference number of the dictionary Dpx becomes n=ll. It is incremented by one (+1). Further, as described above, in order to select the dictionary for the next character string, a new history state PK is obtained by using LUT from 1 on the last character.

Next, the decoding algorithm of the present invention will be explained with reference to FIG.

Decoding is the inverse operation of encoding. First, initialization of the dictionary shown in s100 is similar to encoding.

The procedures from 81 to S8 are basically the same as the conventional method shown in FIG.

The difference in the decoding of the present invention is that after decoding the reference number ω from the input code C0DE in S4, the dictionary DpK is selected using the history state PK obtained from the last character of the previous character string, and the selected dictionary A character string corresponding to the reference number ω is found in the DPK.

Registration of a new character string in the dictionary is similar to the case of LZW encoding, but is performed one tempo later than when encoding. In other words, during encoding, the character string ωK (the 10th character of the character string of interest), which is extended by one character, is registered in the dictionary when the encoding of the character string of interest is completed, but when decoding, the character string of interest is When extending the string ω by one character, it is registered in the dictionary along with the first character of the next string, so the next string must be restored or terminated.

Please register at that time.

Specifically, as shown in S7, the reference number O of the immediately preceding character string
4. Add 1 silk to the first character of the restored character string with L Dω, and select the dictionary selected by the history state PKI from the last character of the immediately previous character string. , -. Therefore, the current history state PK is transferred to PKI in order to extend the restored character string and register it next time, and a new history state is determined by adding 2 to the last character of the restored character string.

In the above embodiment, the dictionary is composed of 16 characters according to the historical state for a total of 256 character types, but if necessary, an appropriate dictionary can be used as long as the total number of character types is less than the total number of character types. The number may be 6, and the number of character types may be determined as necessary.

[Effects of the Invention] As explained above, according to the present invention, it is possible to easily and efficiently register the initial values of multiple dictionaries according to the historical status of character strings.
Since it is possible to incorporate the history of character strings that previously appeared in the character string being encoded, redundancy between character strings is reduced, and a compression ratio higher than that of the conventional 7ZW code can be obtained. is expressed only by a reference number, so it is easy
It can be executed using a suitable algorithm.

[Brief explanation of drawings]

FIG. 1 is a diagram explaining the principle of the present invention. FIG. 2 is a block diagram of an embodiment of the present invention; FIG. 3 is a flowchart of the encoding algorithm of the present invention; FIG. 4 is a flowchart of the decoding algorithm of the present invention. FIG. 5 is a flowchart of a conventional i=zw encoding algorithm. FIG. 6 is a flowchart of the conventional +, ZW decoding algorithm. FIG. 7 is an explanatory diagram of a specific example of conventional LZW encoding. FIG. 8 is an explanatory diagram of a dictionary configuration example. FIG. 9 is an explanatory diagram of a specific example of conventional LZW decoding. FIG. 2 is an explanatory diagram of encoding in which substring decomposition and history between character strings are captured. [? Explanation of No. 1 1 10.10-1-1f')-N, Dictionary 12: cPU 14 Niprogram memory 16 Control software 18 Maximum growth search software 20 Encoding software 22 Decoding software 24 Dictionary initial value creation software 26 Data Memory 28: Data Buffer Patent Applicant Tomi Tsutsu Stock Price'! ]

Claims (10)

[Claims] (1) In a data compression method in which an input character string is specified and encoded using a reference number of a substring that matches the maximum length among already encoded substrings registered in a dictionary (10), the dictionary (10) is composed of a dictionary group consisting of a predetermined number of dictionaries (10-1 to 10-N) smaller than the number of all character types to be processed, and each dictionary (10-1 to 10-N) has at least All character types are initially registered with a reference number attached to each character, and when encoding an input character string, identification in the dictionary group is performed according to index information indicating the dependency relationship with previously encoded character strings. If there is no input string in the specified dictionary (10-i), add the next character to the reference number of the previously encoded string. A data compression method characterized by registering character strings with new reference numbers. (2) In the data compression method according to claim 1, when encoding an input character string, data is stored in the dictionary group according to index information obtained from a part of the final character code of the previously encoded character string. A data compression method characterized by specifying a specific dictionary (10-i). (3) In the data compression method according to claim 2, when encoding an input character string, data is stored in the dictionary group according to index information indicated by the upper bits of the final character code of the previously encoded character string. A data compression method characterized by specifying a specific dictionary (10-i). (4) In the data compression method according to claim 1, when encoding an input character string, the input character string is encoded according to the index information obtained by referring to the lookup table based on the final character code of the character string encoded immediately before. A specific dictionary in a group of dictionaries (1
A data compression method characterized by specifying 0-i). (5) In the data compression method according to claim 4, when encoding an input character string, data is stored in the dictionary group according to index information created from the upper bits of the final character code of the previously encoded character string. A data compression method characterized by specifying a specific dictionary (10-i). (6) Specify the input string with the reference number of the substring that matches the maximum length among the already encoded substrings registered in the dictionary (10), and extract the original string from the encoded code word. In the data restoration method, the dictionary (10) is replaced by a predetermined number of dictionaries (10) that are smaller than the number of all character types to be processed.
-1 to 10-N), each dictionary (1 to 10-N)
At least all character types for each character string (0-1 to 10-N) are initially registered with a reference number attached to each character, and when restoring an input code word, an index is used to indicate the dependency relationship with previously restored character strings. Specify and restore a specific dictionary (10-i) in the dictionary group according to the information, and add the first character of the currently restored character string to the reference number of the previously restored character string for each restoration. A data compression method characterized by registering a character string with a new reference number. (7) In the data restoration method according to claim 6, when restoring an input code word, a specific one in the dictionary group is selected according to index information obtained from a part of the final character code of the character string that has been restored immediately before. A data restoration method characterized by specifying a dictionary (10-i). (8) In the data restoration method according to claim 7, when restoring an input code word, a specific one in the dictionary group is selected according to index information indicated by the upper bits of the final character code of the character string that has been restored immediately before. A data restoration method characterized by specifying a dictionary (10-i). (9) In the data restoration method according to claim 6, when restoring the input code word, the dictionary group is based on the index information obtained by referring to the lookup table using the final character code of the character string that has been restored immediately before. A specific dictionary (10-
A data restoration method characterized by specifying i). (10) In the data restoration method according to claim 9, when restoring the input codeword, the index information obtained by referring to the lookup table is used based on the upper bits of the final character code of the character string that has been restored immediately before. A data restoration method characterized by specifying a specific dictionary (10-i) among the dictionary group.