JP3034016B2 - Data compression and decompression method - Google Patents

Description

Detailed Description of the Invention [Table of Contents] Overview Industrial application field Conventional technology (FIGS. 9 to 13) Problems to be solved by the Invention Means for Solving the Problems (FIG. 1) Action Embodiment (A) Description of the first embodiment (FIGS. 2 to 5) (b) Description of the second embodiment (FIG. 6) (c) Description of the third embodiment (FIG. 7) d) Description of the fourth embodiment (FIG. 8) (e) Description of other embodiments [Summary] Regarding a data compression and decompression method for compressing data by universal encoding, data between irrelevant data sequences In order to prevent data from being registered and to improve the compression ratio, the encoded data is divided into different sub-sequences, and the sub-sequences are registered in the dictionary. Search the column and match the input data with the maximum length of the substrings registered in the dictionary. In a data compression method of coding by specifying a reference number of an object and registering it in the dictionary, local data breaks and global data breaks are detected from the input data, and the data breaks between the local data breaks are detected. ,
A new subsequence obtained by adding the next input data to the matched subsequence is registered in the dictionary, and between the global data delimiters, each of the data from the head data to the final data of the global data delimiter is registered. Subsequences are successively registered in the dictionary.

[Industrial applications]

The present invention relates to a data compression and decompression method for compressing data by universal coding.

In recent years, various types of data such as character codes, vector information, and images have been handled by computers.
The amount of data handled is also rapidly increasing. When dealing with a large amount of data, by compressing the amount of data by omitting redundant portions in the data, it becomes possible to reduce the storage capacity or to transmit data quickly.

Universal coding has been proposed as a method that can compress various data in one system. Here, the field of the present invention can be applied not only to character code compression but also to various types of data. In the following, following the name used in information theory, one word unit of data is called a character, and Are connected to an arbitrary word.

A typical universal code is Ziv-Lemp
There is an el (Jip-Lempel) code (for example, for example,
Munakata "Ziv-Lempel Data Compression", Information Processing, Vol.2
6, No. 1, 1985).

For the Ziv-Lempel code, two algorithms of a universal type and an incremental decomposition type (Incremental parsing) have been proposed. Furthermore, as an improvement of the universal algorithm, there is LZSS code (TCBell, “Better OPM
/1.Text Compression ", IEEE Trans.on Commun., Vol.COM
-34, No. 12, Dec. 1986). As an improvement of the incremental decomposition type algorithm, there is an LZW (Lempel-Ziv-Welch) code (TAWelch, "A Technique for High-Performa").
nce Data Compression ", Computer, June 1984).

In such a data compression method using universal coding, an improvement in compression ratio is required.

[Conventional technology]

 FIG. 13 is an explanatory diagram of the prior art.

Originally, the universal code is an information preservation type data compression method. Since the statistical properties of the information source are not assumed in advance at the time of data compression, it can be applied to data of various types (character codes, object codes, etc.). Can be.

In a document image, there is a similarity in the linearity and the degree of bending of a character line of a character. In a halftone dot image, an enormous number of change points appear because the entire image is halftone dispersed, but the connection relationship of the outlines is similar due to the halftone dot periodicity and the same halftone dot shape. Redundancy having this similarity can be reduced by universal coding, and effective compression can be performed.

As described above, since universal coding is effective for image data, as shown in FIG. 9, preprocessing for converting image data into two-dimensional information (intermediate data) by a pattern run length method or the like is performed. Next, a method has been already proposed in which the present process of performing universal encoding using an LZW code or the like is performed to further increase the compression efficiency of universal encoding for an image.

 The proposed methods include the following.

The universal coding is applied after converting the image data into information of the type of the black and white pattern that is two-dimensionally captured and the number of continuous patterns (pattern run length method).

In order to track the outline of the image data, universal coding is applied after converting the information into information on the connection relationship of changed pixels that change from white to black or from black to white (contour line method).

The input image is converted into the relative address data of the changed pixels, and from the converted data, the statistical properties such as the continuity of the converted pixels in the scanning line direction are learned by the universal coding method to optimize the codes, and various codes are optimized. Efficient compression is performed on an image with the characteristic (modified MMR method).

 This universal encoding will be described using an LZW code as an example.

FIG. 10 is a flowchart of the LZW encoding process, and FIG. 11 is a flowchart of the LZW decoding process.

LZW coding has a rewritable dictionary, divides the input character code and data into different character strings, assigns numbers to these character examples in the order they appear, and registers them in the dictionary.
The currently input character string is represented by the number (index) of the longest matching character string registered in the dictionary and encoded.

 The encoding process will be described with reference to the flowchart of FIG.

First, in step S1 (hereinafter "step" is omitted), a character string consisting of one character for all characters is registered in advance as an initial value, and then encoding is started. In the encoding of S1, the dictionary is searched by the first character K input to obtain a reference number ω, which is used as a prefix string.

Next, at S2, the next character of the input data is read, and at S3, it is checked whether the character input is completed.
A search is made to see if the dictionary has a character (ωK) obtained by adding the character K read in S2 to the initial character string ω obtained in S5 or ω in S5.

If the character string (ωK) is not in the dictionary in S4, the process proceeds to S6 and S1
The reference number ω of the character K obtained in step is output as a code word code (ω), a new reference number is added to the character string (ωK) and registered in the dictionary, and the input character K of S2 is further referred to as the reference number ω And the dictionary address n is incremented, and the process returns to S2 to read the next character K.

On the other hand, if the character string (ωK) is found in the dictionary in S4, the character string (ωK) is replaced with the reference number ω in S5, and the process returns to S2 to search for the maximum matching length until the character string (ωK) cannot be searched from the dictionary. Continue.

The decoding process of FIG. 15 performs the reverse operation of the encoding of FIG.

In the decoding of FIG. 15, similarly to the encoding, the decoding is started after a character string consisting of one character for every character is registered in the dictionary as an initial value in advance.

First, the first code (reference number) is read in S1, and the current code is read.
CODE is OLDcode, and since the first code corresponds to one of the reference numbers of one character already registered in the dictionary, a character code (K) that matches the input code CODE is searched for, and the character K
Is output. Note that the output character (K) is set in FIN char for exception processing in S8 described later.

Then, the process proceeds to S2, where the next code is read and set as CODE in INCODE.

In S3, it is checked whether there is a new code, that is, whether or not the code input has been completed, and the process proceeds to S4, where the code C input in S3
Check whether ODE is defined (registered) in the dictionary.

Normally, since the input code word is registered in the dictionary in the previous processing, the process proceeds to S5, where the character string c corresponding to the code CODE is obtained.
ode (ωK) is read from the dictionary, the character string K is temporarily stacked in S6, the reference number code (ω) is set as a new CODE, and the process returns to S5 again.
Is repeated until one character is reached. Finally, the process proceeds to S7, and the characters stacked in S6 are popped up and output in LIFO (Last In Past Out) format. At the same time, in S7, a new reference number is added to the character string represented as a set (ω, K) of the code ω used last time and the first character K of the character string restored this time, and registered in the dictionary.

A pre-processing method of image data for improving the merit of such universal coding of an image will be described with reference to FIGS. 12 and 13, taking an already proposed method as an example.

Fig. 12 is an explanatory diagram of two-dimensional encoding of image data already proposed,
FIG. 13 is an explanatory diagram of the encoding example.

Taking the example of an 8-line image as shown in FIG. 12 as an example, the proposed pattern run-length method encodes a vertical 8-line pattern and combines it with the horizontal length (run-length) of the same pattern. .

That is, a two-dimensional image is coded by a vertical pattern of eight lines and its horizontal run length (RL).

For example, when white is “0” and black is “1”, and the black and white state of each line is expressed, if eight lines in the vertical direction are white, the pattern code is “00000000” and the length in the horizontal direction is “127”.
Is a run-length code "01111111", which is encoded as one unit.

In the case of FIG. 12, encoding is performed like the input symbol of FIG.

According to this method, an image is encoded as a two-dimensional block, and two-dimensional encoding can be performed by a combination of a pattern and a run length.

When input symbols of such a unit length (byte) are universally encoded (for example, LZW code), the result becomes as shown in FIG. 13, and an output code (maximum length matching code) + the next symbol is registered in the dictionary.

For example, an input symbol “00000000” is encoded with an index “0”, and the next input symbol “01” is stored in the dictionary.
The combination with “111111” is registered as the index “256”.

[Problems to be solved by the invention]

By the way, in the conventional universal coding, data obtained by adding the next one symbol to the maximum length matching data is sequentially registered in the dictionary irrespective of the data sequence.

Therefore, similar to the above-mentioned intermediate data with similarity,
Even if it is data, the registration numbers 257,259,261,263,
A combination of completely unrelated data series such as 265, 267, 269... Is also registered.

For this reason, the following problems have arisen in the prior art.

Since data between irrelevant data series is also registered in combination, a large number of data having an extremely low appearance frequency are registered, so that the learning effect is delayed and the initial compression ratio is reduced.

Index (reference number) because there are many useless registrations
, The number of bits increases, and the compression ratio decreases.

Accordingly, an object of the present invention is to provide a data compression and decompression method capable of preventing data between irrelevant data sequences from being registered and improving the compression ratio.

[Means for solving the problem]

 FIG. 1 is a diagram illustrating the principle of the present invention.

In the present invention, as shown in FIG. 1, the encoded data is divided into different sub-sequences, and the sub-sequences are registered in a dictionary.
Searches a subsequence registered in a dictionary with input data, encodes the input data by specifying a reference number of a subsequence registered in the dictionary that matches the maximum length, and compresses the data into the dictionary Detecting a local data break and a global data break from the input data,
A new subsequence obtained by adding the next input data to the matching subsequence is registered in the dictionary between the local data delimiters, and the global data delimiters are registered between the global data delimiters. Are sequentially registered in the dictionary from the first data to the last data.

The present invention also provides a data restoration method for comparing encoded data with a reference number of a subsequence registered in a dictionary and restoring the encoded data to a subsequence with a matching reference number. A local data segment and a global data segment are detected from the restored subsequence, and a new subsequence obtained by adding the next input data to the matched subsequence is detected between the local data segments. The sub-strings are registered in the dictionary, and the sub-sequences from the leading data to the last data of the global data break are continuously registered in the dictionary between the global data breaks.

[Action]

According to the present invention, dictionary registration is performed using data between local data delimiters, so that a combination of data having different local data series described above is prevented from being registered, and unnecessary registration is prevented. To promote.

On the other hand, looking at the data globally, there is a global data sequence such as an image group.

When divided by the above-mentioned local data series, the number of registered combination symbols is limited to the interval, and does not become long.

Therefore, the data is divided even in the global data series, and the data during this period is registered consecutively from the first symbol so that the combination of the data between the different local data series described above does not occur. Is to improve.

〔Example〕

(A) Description of the first embodiment FIG. 2 is a processing flowchart of the first embodiment of the present invention.

As described in FIGS. 16 and 17,
Convert to two-dimensional information and preprocess.

Next, global and local breaks (breaks) of the converted data are recognized (detected), and the above-described universal encoding is performed while the dictionary is selected for registration.

 This processing is performed as follows.

The same character (data) sequence as the input character (data) sequence is searched in the dictionary (dictionary search step).

Next, encoding is performed using an index of a character (data) series that has been searched and matched in the dictionary (index encoding step).

Recognize global and local breaks in a character (data) string and register them in a dictionary (dictionary registration step).

FIG. 3 is a flowchart of an encoding process according to the first embodiment of the present invention, FIG. 4 is a flowchart for explaining the number of operations, and FIG. 5 is a flowchart of the decoding process.

It is assumed that a processor (not shown) creates and executes a dictionary in a memory for both encoding and decoding.

 The encoding process will be described with reference to FIG.

The same steps as those shown in FIG. 10 are denoted by the same symbols.

S1) To initialize the dictionary, the first 256 characters (words) are registered, and the start address N of the dictionary is set to 256.

Next, the first character K is input, and its address is assigned to the registered character string ω ₁ and the code character string ω ₂ in the dictionary.

S2) Input the next character K and proceed to S4.

S4) a combination of a string ω ₁ and the letter K determine whether there is in the dictionary.

S5) If ω ₁ K exists in the dictionary, substitute the registered address of ω ₁ K into ω ₁ and the registered address of ω ₂ K into ω ₂ .

S3) Next, it is checked whether the data is completed.
If not, return to S2.

Thus, the longest matching registered character string in the dictionary is searched.

If S7) data end, and outputs an address omega ₂ as a code, ends.

S8) In step S4, if ω ₁ K does not exist in the dictionary, it is determined whether or not ω ₁ and K are global breaks in data.

For example, when an input symbol (intermediate data) is coded by a pattern run length, a global break (break) distinguishes a pattern including black from a white pattern, as shown in FIG. It can be analyzed by decoding the code.

S9) If data is global break, outputs address omega ₂ as a code without registration, ₁ dictionary address of K omega, omega
_Then , the process proceeds to step S3.

When a global break is detected in this way, ω ₁ K between the breaks is not registered, and the symbol K following the break is changed to ω ₁ , ω ₂
And start the search registration from K.

S10) On the other hand, the data determines whether if not global cut, then the combination of the code strings omega ₂ and the character K is in the dictionary.

S11) If ω ₂ K is in the dictionary, register ω ₁ K in the dictionary,
A character string from the head of the global division to the character K is registered (referred to as continuous registration), the reference address N is incremented to N + 1, and the dictionary address of ω ₁ K is substituted for ω ₁ .

S12) Next, it is determined whether or not data between ω ₂ K is a local break.

The local breaks in the data are set as pairs of patterns and run lengths, as shown by, for example, ↓ in FIG.

S13) If it is not a local break, the dictionary address of ω ₂ K is substituted for ω ₂ for registration between local breaks, and the process proceeds to step S3.

S14) On the other hand, if the local cuts, and outputs the address omega ₂ as the code, since registration is done at S11, in order to the next character K of the cut at the beginning of the code string, a dictionary address of the character K omega Set to ₂ and go to step S3.

S15) In S10, if ω ₂ K is not in the dictionary, address ω ₂
Is output as a code, ω ₁ K is registered in the dictionary, and the reference address N is incremented to N + 1 as in S11.

S16) Next, it is determined whether ω ₂ = ω _1, if ω ₂ = ω _1, since there is no need registration omega ₂ K, the process proceeds to step S18.

S17) If ω ₂ ≠ ω _1, for the registration of character strings between local separated, the omega ₂ K to dictionary registration, the reference address N N +
The value is incremented to 1 and the process proceeds to step S18.

S18) in order to extend the registration string omega _1, substitutes the registered address of the omega ₁ K to omega _1, the code string omega ₂ substitutes the dictionary address character K, the process proceeds to step S3.

 This will be specifically described with reference to FIG.

The input symbols shown in FIG. 4 are those obtained by performing the above-described pattern run-length two-dimensional coding, and are obtained by coding a set of a pattern and a run length.

Local breaks are set for each set, and global breaks are set by distinguishing white patterns from other patterns including black.

In FIG. 4, a local delimiter is set every two bytes, and a global delimiter is set by a delimiter between a white pattern and other black patterns.

This is because a series of patterns including black is considered as one image.

If it is not a global break, ie, between global breaks, S8,
By setting the _first character K of the break to ω1 as in S9, the _first character K starts with S11 and S15, the registration number “258”,
Continuous registration is performed as "259", "261", "262", "264", "265", and the like.

On the other hand, if the local cut, by the letter K and omega ₂ at S14, the normal registration data between cuts in S17 is registration number "257" is carried out such as "260".

With the registration number "265", a data sequence that is globally significant as an image is registered as a collective data sequence, so that compression can be performed efficiently.

In addition, since registration is not performed so as to straddle local delimiters, useless registration can be prevented.

 Next, the decoding process will be described with reference to FIG.

The same steps as those shown in FIG. 11 are denoted by the same symbols.

S1) As in the case of the encoding, the first 256 characters are registered in the dictionary in advance, and decoding is started after the head address n of the dictionary is set to “256”.

First, the first code (reference number) is read and the current CO
DE is an OLD code. Since the first code corresponds to one of the reference numbers of one character already registered in the dictionary, a character code (K) that matches the input code CODE is searched for and the character K is output. Note that the output character (K) is set in FIN char for later exception processing.

S2) Next, proceed to S2, read the next code, and set INCO to CODE.
Set as de.

S4) Next, proceed to S4, and check whether or not the code CODE input in S2 is defined (registered) in the dictionary.

S5) Normally, the input codeword has been registered in the dictionary in the previous processing, so the process proceeds to S5, where the character string code (ω ₁ K) corresponding to the code CODE is read from the dictionary.

S6) Character string K is temporarily stacked in S6, and reference number co
de (ω ₁ ) is returned to S5 as a new CODE, and S5, S5
Step 6 is recursively repeated until the reference number ω1 reaches _one character.

S9) Finally, proceed to S9, output the character K, and set K to FINchar
LIFO (Last In F)
ast Out) pop up and output.

S10) Next, the restored character is parsed in the same manner as in the above-described encoding step S8 to check whether there is a global break in the data.

If the data is a global break, the process proceeds to step S11. If the data is not a global break, the process proceeds to step S12.

S11) If the data is a global break, the dictionary is not registered.
Set the INcode to OLD code, to sign string ω ₂ "0"
Is set, and the process proceeds to step S3.

S3) Check whether the data is completed.
Return to step 2 and end the decoding if done.

S12) If the data is not a global break, determine whether it is a local break.

If it is a local break, the process proceeds to step S13. If it is not a local break, the process proceeds to step S14.

If S13) local cuts, in order to letters K to the head of the code string omega _2, and substituted into omega _2, the process proceeds to step S17.

S14) If it is not a local break, determine whether ω ₂ K is in the dictionary.

S15) If not in the dictionary, since it has not been registered, ω ₂ K is registered in the dictionary, and the reference address n is incremented to n + 1.
Proceed to step S16.

 As a result, normal registration between local breaks is performed.

S16) Next, the dictionary address of ω ₂ K is substituted for ω ₂ .

S17) In S17, a character string representing a pair (OLD code, K) of the code OLD code used last time and the first character K of the character string restored this time is registered in the dictionary with a new reference number, and the reference number ( Address) n is incremented to n + 1 and OLD code
Set the character of the combination of K and K to OLD code and step
Go to S3.

S8) If the code is not registered in S4 (it occurs when referring to the immediately preceding reference number in encoding), FIN char is output in S8, OLD code is set to CODE, and code
After returning (OLD code, FIN char) to IN code, proceed to S5.

 In this way, decoding is performed in the same way as encoding.

(B) Description of Second Embodiment FIG. 6 is an explanatory diagram of a second embodiment of the present invention.

In this example, the contour method is incorporated in the preprocessing, and is shown in the image example of FIG.

In the contour method, the contour of the image is represented by a horizontal mode and a run length RL on one line, and the other lines are sequentially represented by a horizontal displacement length ZL from the line.

For example, the left outline of the left black image in FIG.
In the figure, it is represented as contour 1 and the right contour is contour 2
Is represented by

That is, the contour line 1 has a boundary of the first line at the 128th address from the left end of the first line, the second line is shifted from the first line by “0” (ZL0), and the third line is Offset "1" (ZL-1) with respect to the line, 4th line with "0" (ZL0) with respect to the 3rd line, white after 5th line (pass mode P)
It is expressed as

Here, in Fig. 6, the local data delimiters are ↓,
Global breaks are indicated by. The general division in this case is based on a data group indicating each contour line (codes from H to P) as a guide, and is considered to form one image contour. In addition, as a local delimiter, the horizontal code was conscious of the distinction between a vertical code and a pass code, and it was thought that the horizontal code could capture the horizontal correlation and the vertical and pass code could capture the vertical correlation. It should be noted that the line number where each contour line exists and the contour line number adjacent to the line number are only added as output codes, and are not registered in the dictionary here. (In order to prevent an increase in the index) For example, for the contour line 1, a pass code is registered in the dictionary as a series of image data through three vertical codes from a horizontal code. (Registration number 264) Therefore, a data sequence that is globally significant is registered as an image, so that compression can be performed efficiently.

(C) Description of Third Embodiment FIG. 7 is an explanatory view of a third embodiment of the present invention.

In this example, in the contour method shown in FIG. 6, a mode separation type in which the horizontal, vertical, and pass modes of each contour and the run length are separated is shown.

Here, in Fig. 7, the local data delimiters are ↓,
Global breaks are indicated by. In this case, as a global delimiter, an image contour is formed by dividing a data group (codes from H to P) indicating each contour line into a mode code and a run-length code. Regarding local breaks, only run-length codes were distinguished from horizontal codes, vertical codes, and pass codes.

Now, for example, in the outline 1, the mode codes are collectively registered in the dictionary (registration number 260), so that the subsequent outlines 2, 3, and 4 can be represented by the registration number 260. Therefore, since a data sequence that is globally significant as an image is registered as a collective data sequence, it can be efficiently compressed.

(D) Description of Fourth Embodiment FIG. 8 is a processing flowchart of the fourth embodiment of the present invention.

In this embodiment, as compared with the embodiment of FIG. 2, when converting image data into two-dimensional information, a break (break) code is inserted at a global and local break position. This is to detect the delimiter code and perform the encoding and decoding described in the first to third embodiments.

By doing so, there is an advantage in that the parsing does not have to be performed for the detection of the break, and the encoding and decoding times can be reduced.

(E) Description of Other Embodiments In addition to the above-described embodiments, the present invention can be modified as follows.

Although image data is targeted, other data may be used.

Although image data is converted into two-dimensional information and encoded, original data may be encoded.

Although the present invention has been described with reference to the embodiments, the present invention can be variously modified in accordance with the gist of the present invention, and these are not excluded from the present invention.

〔The invention's effect〕

As described above, the present invention has the following effects.

Since dictionary registration is performed using data between local data breaks, it is possible to prevent a combination of data having different local data series from being registered, prevent unnecessary registration, and promote a learning effect.

Looking at the data globally, there is a global data sequence such as an image group. However, if the data is divided by the above-mentioned local data sequence, the number of registered combination symbols is limited to the interval,
It will not be long.

Therefore, the data is divided even in the global data series, and the data during this period is registered consecutively from the first symbol so that the combination of the data between the different local data series described above does not occur. Improve.

[Brief description of the drawings]

 FIG. 1 is a principle diagram of the present invention, FIG. 2 is a processing flowchart of the first embodiment of the present invention, FIG. 3 is a flowchart of an encoding processing of the first embodiment of the present invention, and FIG. FIG. 5 is an explanatory diagram of the operation of the first embodiment, FIG. 5 is a flowchart of the decoding process of the first embodiment of the present invention, FIG. 6 is an explanatory diagram of the second embodiment of the present invention, and FIG. FIG. 8 is an explanatory view of a third embodiment, FIG. 8 is a processing flowchart of a fourth embodiment of the present invention, and FIGS.

────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Yasuhiko Nakano 1015 Uedanaka, Nakahara-ku, Kawasaki City, Kanagawa Prefecture Inside Fujitsu Limited (56) References JP-A-60-116228 (JP, A) JP-A-61-242122 (JP, A) (58) Field surveyed (Int. Cl. ⁷ , DB name) H03M 7/42

Claims (2)

(57) [Claims] 1. A method according to claim 1, wherein the encoded data is divided into different subsequences, the subsequences are registered in a dictionary, the input data is searched for the subsequences registered in the dictionary, and the input data is registered in the dictionary. In a data compression method of coding by specifying a reference number of a subsequence that matches the maximum length among reference subsequences and registering it in the dictionary, a local data delimiter and a global data delimiter are detected from the input data. A new subsequence obtained by adding the next input data to the matched subsequence is registered in the dictionary between the local data delimiters, and the global data is delimited between the global data delimiters. A data compression method characterized by successively registering, in the dictionary, each subsequence from the first data to the last data of the delimiter. 2. A data restoration method according to claim 1, wherein said encoded data is compared with a reference number of a subsequence registered in a dictionary, and said encoded data is restored to a subsequence having a matching reference number. A local data segment and a global data segment are detected from the divided subsequence, and a new subsequence obtained by adding the next input data to the matched subsequence is detected between the local data segments. Registering in the dictionary, and successively registering, in the dictionary, each subsequence from the first data to the last data of the global data break between the global data breaks. Method.