JP2614318B2 - Halftone image data compression method - Google Patents

Description

[Contents] Outline Industrial application field Conventional technology (FIGS. 8 and 9) Problems to be Solved by the Invention Means for Solving the Problems (FIG. 1) Action Embodiment (1) Universal type algorithm (FIG. 2) (2) Incremental decomposition type algorithm (FIG. 3) (3) First embodiment (FIGS. 4 and 5) (4) Second embodiment (FIG. (FIGS. 6 and 7) Effects of the Invention [Overview] Regarding a halftone dot image data compression method, an object of the present invention is to enable efficient compression with a small hardware configuration, and to hold an image data to be compressed and encoded. Data holding means, preceding data holding means for holding data subjected to compression code processing, comparing means, mismatch number counting means for counting the number of mismatches, and detecting whether or not the number of mismatches is within a specified value Threshold means and the maximum matching length detection A matching length detection means comprises a coding means, by sequentially comparing the input image data held in the input data holding means by said comparing means and data held in said preceding data holding means,
The maximum number of input data subsequences whose number of mismatches is within a specified value is collectively compressed and encoded by the encoding means.

[Industrial applications]

The present invention relates to a halftone dot image compression method, and more particularly to a compression method using a Ziv-Lempel code.

In recent years, office automation has been developed, and gradation images have been handled in addition to characters as document information. Since a document is usually represented by binary values of white and black, in order to represent a gradation image, an area modulation method in which the number of black pixels among halftone dots composed of a plurality of pixels is determined according to the gradation. It is common to express using. When document information is used as digital data, the data amount of a gradation image is much larger than that of a character image, and is several ten to several tens times. Therefore, it is necessary to reduce the amount of data by applying efficient data compression in order to efficiently handle a gradation image in storage or transmission.

[Conventional technology]

As a conventional international standard compression method for binary images, there is a modified read (MR) method. In this MR method, a pixel that changes from white to black or from black to white when viewed in the main scanning direction is called a change point, and a change in the boundary of the black and white pattern represented by the change point between adjacent scanning lines is small. Data compression is performed by focusing on the connection relationship between points.

However, in a halftone image that has become a halftone dot, the number of change points caused by the halftone dot dispersed over the entire screen is enormous, so that the MR method cannot be effectively compressed.

For this reason, as a method of data compression of a halftone image, a predictive coding method has been used separately from the MR method which is the standard method.

As shown in FIG. 8, this method takes a reference pixel rd around the target pixel do and a position away from the halftone dot period, predicts white and black of the target pixel do, and encodes a prediction error. Was used. However, when a document is read by a scanner and data is compressed, as in a facsimile, the types of halftone dots are various and the halftone period is often unknown in advance.
Therefore, in the related art, an adaptive predictive coding method in which a plurality of predictors having various halftone dot periods as reference pixels are arranged, a predictor with the least number of prediction errors is selected, and coding is performed according to the selected predictor. I was using.

As shown in FIG. 9, the transmitted halftone image signal is input to a line memory 50, and the output thereof is input to a first predictor 51 and a second predictor 52 having different halftone periods, respectively, to obtain respective predicted values. The exclusive OR circuits 53 and 54 detect whether or not these predicted values match the actual halftone image signal, and count the number of mismatches with the first prediction error counter 55 and the second prediction Counting is performed by the off counter 56. The counting result is compared by the comparison circuit 57 for each fixed number of input halftone image signals, and in the next section, a predictor having less deviation is selected by the multiplexer 58, and the predictor having the less prediction deviation is selected from the predictor having the smaller prediction deviation. The prediction error signal was compression-encoded by the encoder 59. Therefore, if the prediction in the selected predictor matches the input halftone image signal, there are many “0” in the prediction error signal output from the multiplexer 58,
Since the number of “1” is small, the compression efficiency in the encoder 59 is improved. The conventional adaptive prediction coding method of the prior art is described in detail in, for example, the IEICE technical report IE80-01 "Adaptive prediction coding of newspaper halftone photographs".

[Problems to be solved by the invention]

However, in the conventional halftone image data compression method as described above, in order to configure a predictor in anticipation of the statistical properties of the halftone image, the halftone dot period of the prepared predictor is set to the actual value. Although effective data compression can be performed when the halftone dot period matches the halftone period of the image, there is a problem that when the halftone dot period does not match, the data compression efficiency is significantly deteriorated.

By adopting such an adaptive predictive coding method, the problem of reduced data compression efficiency due to the dot period mismatch can be improved to some extent. Since it is necessary to increase the number of the circuits, there is a problem that the circuit scale becomes large.

Accordingly, an object of the present invention is to provide a halftone image data compression system capable of appropriately compressing data at various periods and compressing data with a small circuit scale in order to improve such a problem.

[Means for solving the problem]

In order to achieve the above object, as shown in FIG. 1, the present invention relates to a buffer memory 1, a preceding data memory 2, an exclusive OR circuit 3, a first address control unit 4, a second address control unit 5, It includes a counter 6, a threshold circuit 7, a maximum matching length detection unit 8, an encoding unit 9, and the like.

Here, as described in detail in an embodiment to be described later, when the universal type Jib-Lempel code is used, the precedent data which has been subjected to the encoding processing is sequentially written in the preceding data memory 2, When the code is used, the existing component sequence created by the preceding data is entered.

Further, as described later in detail, when the universal type Jib-Lempel code is used, the encoding unit 9 determines, in the preceding data memory 2, the address of the matching start point where a substantially matching pattern exists, and the pattern length. And the next symbol is output as a compression code, and when the incremental decomposition type Jib-Lempel code is used, the index of the coding pattern component which substantially matches in the preceding data memory 2 and the next symbol are output as the compression code. .

When the halftone image signal is input to the buffer memory 1, the first address control unit 4 first outputs the first one symbol to the exclusive OR circuit 3, and the second address control unit 5 outputs the first symbol from the preceding data memory 2. The held values are sequentially output, and the equal values are detected. When the equality is detected in this way, the first address control unit 4 outputs the second symbol to the exclusive OR circuit 3. At this time, the second address control unit 5 outputs the next held value that has previously been matched. As a result, if a further match is obtained, the first address control unit 4
Outputs the third symbol to the exclusive OR circuit 3,
Further, the second address control unit 5 outputs the next hold value and the coincidence two consecutive times. However, if this third mismatch does not occur, the exclusive OR circuit 3 outputs "1" and the mismatch counter 6 counts the number of mismatches. Then the first
The address control unit 4 outputs the fourth symbol to the exclusive OR circuit 3, and the second address control unit 5 outputs the next data that does not match. If the fourth matches again, the first address control unit 4 and the second address control unit 5 compare the fifth symbol. At this time, if there is a mismatch again, the mismatch counter counts 2. If the threshold value Th of the threshold circuit 7 is set to a numerical value “1”, when the mismatch counter 6 counts up 2, the threshold circuit 7
Outputs this mismatch signal to the first address controller 4 and the second address controller 5. As a result, the first address control unit 4 and the second address control unit 5 stop the match detection processing. The maximum length (4 symbols in this example) until the mismatch signal is output from the threshold circuit 7 to the maximum match length detection unit 8 and the mismatch signal are transmitted to the symbol coding unit 9 to which the mismatch signal was output. Outputs a compression code.

[Action]

Even if there is a mismatch within the range set in advance by the threshold circuit, similar patterns can be compression-encoded in as long a state as possible, so that dot image data can be efficiently compressed.

〔Example〕

In the conventional method, as described with reference to FIGS. 8 and 9, the predictor is constructed from the predicted sample images by predicting the statistical properties of the images. In such a case, the compression efficiency becomes poor.

On the other hand, the present invention adopts a universal code method for optimizing a code while learning the statistical properties of the image in the process of compressing the input image, and efficiently compresses various halftone images. Is performed.

Originally, the universal code is an information preservation type data compression method, and does not assume in advance the statistical nature of the information source at the time of data compression, so that it can be applied to data of various types (character codes, object codes, etc.). it can. By the way, in a halftone image, an enormous number of change points appear because the entire image is halftone dispersed. However, the tendency of the change point due to the dot periodicity and the same dot shape has a global regularity when summarized, so the redundancy with this regularity can be reduced by applying the universal code. Reduction and effective compression can be performed.

The present invention applies universal coding to image data, finds a difference between an already-encoded sequence and a current-encoded sequence, and approximates a sequence having a small difference by an already-encoded sequence, thereby saving information. The encoding is performed to obtain a high compression ratio.

Before describing the present invention in detail based on embodiments, a universal code will be briefly described. As a typical method of the universal code, Ziv-Lempe
l) There is a code (for details, for example, see Munakata (Ziv-Lempel Data Compression Method), Information Processing, Vol. 26, No. 1, 1985.) In the case of Jib-Lempel code, (1) Universal Type and (2) Incremental parsing
Two algorithms have been proposed and will be described.

(1) Universal type algorithm This algorithm requires a large amount of calculation but can obtain a high compression rate. This is a method in which coded data is divided into a maximum-length sequence that matches from an arbitrary position of a past data sequence, that is, a subsequence, and encoded as a copy of the past sequence. FIG. 2 is a diagram showing the principle of creating a universal Jib-Lempel code. The P buffer stores encoded input data, and the Q buffer receives data to be encoded. Input to Q buffer as shown in FIG.
When C1, C2, C3, C4, C5... Are inputted, a data sequence in a P buffer in which encoded input data is stored is searched, and these inputs C1,
Find the substring of the maximum length that matches C2. in this case,
Substrings that match the input data C1, C2, C3, C4 are detected. To specify this maximum subsequence sensed in this P buffer, a match starting position p ₁ of the position from the beginning in the P buffer, the maximum length length of the matched and q _1, to match the input data With the next symbol of the subsequence, encoding of a set as shown in FIG. 2 (B) is performed. In this example, "p _1,
q _1, C5 "is a jib - the Lempel codes. Next, the coded sequence in the Q buffer is transferred to the P buffer, and the same processing is performed on unprocessed data of C6 and lower. Such an operation is repeated, and the input data is decomposed into sub-sequences and encoded. FIG. 2C is a block diagram showing this processing. In the above description, the input data has been described with an example of an 8-bit width. However, the present invention is not limited to this, and it is possible to use a symbol larger than this or a symbol as small as 4 bits or 1 bit. Note that P
If the length of the buffer and 4k, P ₁ can be a 12-bit.

(2) Incremental decomposition type algorithm This algorithm is inferior to the universal type of (1) in compression rate, but is simple and easy to calculate. The incremental decomposition type Jib-Lempel code is a sequence of input symbols x
If x = aabababaa..., The incremental decomposition into the component sequence x = X ₀ X ₁ X ₂ ... Is performed as follows.

When the input symbol sequence x is rewritten in the longest column in which the next symbol is combined with the existing component of the input symbol, the result is as follows.

x = a · ab · aba · b · aa (1) In the above equation (1), the first “a” has no a before it, so a is added to an empty column, and the next “ab” Is "a"
Has been added as a component, and the next symbol “b” has been added to this, and “aba” has been added with the next symbol “a” since “ab” has already existed as a component. In this way, when Xj is the longest column from which the rightmost symbol of the existing component has been removed, the component sequence is expressed as follows.

X ₀ = lambda (empty _{sequence), X 1 = X 0 a} , X 2 = X 1 b, X 3 = X 2 a, X 4 = X 0 b, X 5
= X ₁ a ... For example (1) "ab" in the formula, a is because X ₂ = X ₁ b is represented by X ₀ a. Note that j of Xj indicates a component index. Then, as shown in FIG. 3, encoding is performed using a combination of a component index and the next symbol. Hence the X ₃ is a X ₃ = X ₂ a, is coded in the form of "2a". As described above, the incremental decomposition type algorithm obtains a code string having the maximum length matching from the subsequences decomposed in the past, and encodes them as a copy of the subsequences decomposed in the past.

By the way, when such an encoding system is used, the compression efficiency is improved as long as a long symbol sequence matches. The basic idea of the present invention is to detect a symbol sequence that is substantially the same as a subsequence that is as long as possible without being subject to slight differences.

In other words, when an image that has already been halftone-dotted is read digitally, quantization noise is introduced into each pixel at the time of reading, so that even if the same tone continues, it is reproduced exactly as a row of halftone dots of the same size. Never. Therefore 1
The importance of image quality at the pixel level is weakened, and the gradation of the pixel density is expressed.Therefore, it is not always necessary to express the image by complete information preservation type coding, as long as the image quality is not adversely affected. It is important to increase the compression ratio by the information non-storage type. The present invention in view of this point, when comparing the data to be coded and the current coded sequence,
The number of discrepancies is counted, and when the discrepancy count value is equal to or less than a predetermined threshold value, it is regarded as the same sequence and treated in the same manner as a match. Select a partial example of

Hereinafter, the present invention will be described in detail based on a first embodiment and a second embodiment.

(3) First Embodiment A first embodiment of the present invention will be described with reference to FIGS. FIG. 4 shows an embodiment in which the universal type Jib-Lempel code described in the above (1) is used.
The figure is a detailed explanatory view of the main part. In FIG. 4, 1
0 and 11 are buffer memories respectively, 12 and 13 are address controllers, 14 and 15 are registers, 16 is an exclusive OR circuit, 17 is a shift register, 18 is a mismatch counter, 19 is a threshold circuit, and 20 is Denotes a maximum matching length detection unit, and 21 denotes an encoding unit.

Buffer memory 10, which corresponds to the buffer memory 1 in FIG. 1, which dot image data from the input terminal T ₁ is inputted, the buffer memory 11 to the input halftone image data processed coded The address control unit 12 reads unprocessed halftone image data held in the buffer memory 10 or writes new halftone image data, which corresponds to the preceding data memory 2 in FIG. The address controller 13 corresponds to the first address controller 4 shown in FIG. 1. The address controller 13 reads the halftone image data held in the buffer memory 11 and writes the data sent from the buffer memory 10. This corresponds to the second address control unit 5 in FIG.

The register 14 holds the data output from the buffer memory 10, the register 15 holds the data output from the buffer memory 11, and the exclusive OR circuit 16 is stored in the registers 14 and 15. This is for comparing whether or not the data matches, and corresponds to the exclusive OR circuit 3 in FIG. 1, and the shift register 17 sets the comparison result of the exclusive OR circuit 16.

The mismatch counter 18 counts "1" indicating the mismatch transmitted from the shift register 17, that is, the number of mismatches in the exclusive OR circuit 16, and corresponds to the mismatch counter 6 in FIG. Is for detecting whether or not the count value of the mismatch number counter 18 is within a threshold Th, and corresponds to the threshold circuit 7 in FIG.
The maximum match length detection unit 20 holds the start position and the number of words in the buffer memory 11 of the maximum length subsequence that matches within the range of the difference number of the threshold value Th from the subsequence in the buffer memory 10 and the content of the next word. This corresponds to the maximum matching length detection unit 8 in FIG. The encoding unit 21 includes a first
This corresponds to the encoding unit 9 in the figure, and a code as shown in FIG. 2 (B) is represented by the start position and the number of words and the next symbol of the maximum length subsequence transmitted from the maximum matching length detection unit 20. This code is variable-length coded by a known method such as Huffman code or arithmetic code, and output terminal T ₂
It is output as a compression code.

 Next, the operation of FIG. 4 will be described.

Halftone image data from the input terminal T ₁ is inputted and stored in the buffer memory 10. In the halftone image data in the buffer memory 10, the encoded words are sequentially stored in the buffer memory 11.
And the dot image data for the same word is newly stored in the buffer memory 10. Here, one word of the buffer memories 10 and 11 is set to an arbitrary number of pixels equal to or more than one pixel. Therefore, as the number of pixels per word increases, the probability of mismatching increases, and the compression rate decreases, but the processing speed increases because the processing unit increases.

The data held in the buffer memories 10 and 11 are read out as follows, and collated in the exclusive OR circuit 16.

Now, the q _1, q ₂ ...... qn data of each word from the beginning of the buffer memory 10, and p _1, p ₂ ...... pm data of each word in the buffer memory 11 from the beginning. Matching, first reads the first data q ₁ of the buffer memory 10 by the address control unit 12, is set in the register 14. Then, the data p ₁ is read from the head of the buffer memory 11 by the address control unit 13 and set in the register 15. Then, these are collated by the exclusive OR circuit 16 for each pixel, and the collation results are output in parallel and set in the shift register 17. In this case, of course, the exclusive OR circuit 16 has a capacity to compare the data of the registers 14 and 15 at the same time. In order to count the number of mismatches in the collation result set in parallel in the shift register 17 as described above, the shift register 17 is shifted, and the number of “1” in the serial output is counted by the mismatch counter 18. In this way, the mismatch counter 18 counts the number of mismatch bits at the time of comparison. This count value is checked by the threshold circuit 19. As a result of comparing q ₁ and p ₁ , when the mismatch counter 18 counts one more than the threshold Th,
The mismatch signal is output from the threshold circuit 19 to the address control unit 13. As a result, the address control unit 13
Reset the sets the next data p ₂ from the buffer memory 11 to the register 15, the same verification is performed, which in the case of disagreement, the collation with the next data p ₃ is performed. In this way, when q ₁ does not all match the data p ₁ , p _2, ... Pm in the buffer memory 11, the term of the matching length in FIG. It is shown that there is no similar pattern, and data in which q ₁ is written in the next symbol term is created by the encoding unit 21 as a compression code,
The compressed code is, for example, is compressed by the Huffman method is output from the output terminal T _2.

Incidentally, when such a collation, for example, q ₁ is when clogging substantially aligned coincident with q ₂ and the threshold range, the threshold circuit 19 reports this to the address control section 12 and 13. Thus, the maximum matching length detection unit 20 holds the position and the data length of the p _2, The address control unit 12 to the register 14 of q ₁ of the following
outputs q _2, the address control unit 13 to the register 15 of p ₂ follows
Outputs p _3, performs matching q ₂ and p _3. The number of mismatches of this collation is added to the count value of the mismatch number counter 18, but if it exceeds the threshold Th, this is notified to the address control units 12, 13, and
Address control unit 12 to the register 14, first outputs q _1.
Together with the time address control unit 13 resets the mismatch counter 18, first output p ₃ in the register 15, performs verification. If a mismatch, the address control unit 13 in the same manner to output to match the following p _4. Thus, when q ₁ and p ₄ are roughly aligned, the address control section 12 reads the q _2, the address control unit 13 collates reads p _5. When Accordingly schematic match, maximum matching length detection unit 20 is, in turn, its position p ₄ and the number of words holds (2 in this example), this time the address control section 12 reads the q _3, the address control unit 13 is collated reads p ₆ is performed. This is done sequentially,
The start position and the number of words of the maximum match length data held by the maximum match length detection unit 20 are sent to the encoding unit 21, and
Is sent to the next symbol that is held thereto also encoding unit 21, the output terminal T ₂ is the first view these such compressed code shown in (B) is created which further example is compressed by Huffman coding Is output.

In this way, a sequence obtained by compressing and encoding the data at the head of the buffer memory 10 is transferred to the buffer memory 11,
Input the same number of image data from the newly input terminal T ₁ to the buffer memory 10. Such an operation is repeated, and the data is decomposed into partial strings and sequentially compression-encoded.

The address control unit 12, the address control unit 13, and the maximum matching length detection circuit will be described in further detail based on the circuit configuration example shown in FIG.

Image data is input from the terminal T _1, the buffer memory 10
The data in the buffer memory 10 is newly inputted and accumulated by the number of encoded data, and at the same time, the encoded data is moved to the buffer memory 11. The capacity of the buffer memory 10 is m words, and the capacity of the buffer memory 13 is n words. Data storage and reading to and from the buffer memory 10 are controlled by an address control unit 12, and data storage and reading to and from the buffer memory 11 are controlled by an address control unit 13. The head address of the uncoded data in the buffer memory 10 is the counter 12-1
The last address of the uncoded data in the buffer memory 10 is stored in the counter 12-3. In the buffer memory 11, the start address and the end address of n words of the encoded data are stored in counters 13-1 and 13-4, respectively. Counter 20−
In 2, the number of encoded data is set for each successive encoding. The transfer of data to the buffer memory 10 and the buffer memory 11 is performed as follows. Counter 20
Each time -2 is counted down by one, the counter 12-
The data at the address 1 is read from the buffer memory 10 and written to the address of the counter 13-4. Then, after the counter 12-1,12-3,13-1,13-4 counted up one, and writes the data inputted from the terminal T ₁ to the address counter 12-3 of the buffer memory 10.

Encoding of data in the buffer memory 10 is performed as follows.

The head address of the buffer memory 13 of the counter 13-1 is set in the counter 13-2.

The contents of the counter 13-2 are set in the counter 13-3.

The contents of the counter 12-1 are set in the counter 12-2.

The contents of the address of the counter 13-3 are read from the buffer memory 11.

The contents of the address of the counter 12-2 are read from the buffer memory 10.

The contents read from both buffer memories are compared, and the magnitude of the difference is checked by the threshold circuit 19.

If the difference is equal to or smaller than the threshold value by the threshold circuit 19, the counters 12-2 and 13-3 are incremented by one, and the sequence returns to the main routine.

If the difference is greater than the threshold value as checked by the threshold circuit 19, first, the counters 13-2 and 13-
3 is subtracted to determine the match length. Then, the counter 20-2 storing the maximum matching length up to the previous time and the comparator
Compare the magnitudes according to 20-1. If the match length obtained this time is shorter than the match length up to the previous time, the process goes to. Also,
If the match length found this time is longer than the match length up to the previous time,
The match length obtained this time is stored in the counter 20-2. At the same time, the contents of the counter 13-2, which is the start position of the character string, are stored in the register 20-3. Furthermore, register 20-4
From the register 14 are stored. The values of the counter 20-2 and the registers 20-3 and 20-4 are coded by the coding unit 21 by scanning the buffer memory 11 and finding the maximum matching length.

To inspect the next character string in the buffer memory 11,
The counter 12-2 is incremented by one. If the value of the counter 13-2 has not reached the final address of the counter 13-4, the process returns to the step.

From the above operations, the maximum match detection circuit
13 finds the maximum match length. When the maximum match length is obtained by scanning the entire buffer memory 11, the counter 20-2,
The values of the registers 20-3 and 20-4 are passed to the encoding unit 21 and encoded.

When the buffer memories 10 and 11 are 1 bit / word, the shift register 17 becomes unnecessary and the encoding of the next word of the maximum matching subsequence becomes unnecessary.

(4) Second Embodiment A second embodiment of the present invention will be described with reference to FIGS. FIG. 5 shows an embodiment in which the Jib-Lempel code of the incremental decomposition type described in the above (2) is used, and FIG. 7 is a diagram for explaining the operation thereof.

In FIG. 6, reference numeral 30 denotes a buffer memory, which is an input terminal.
The halftone image data is input from T1 and corresponds to the buffer memory 1 in FIG. _1. Reference numeral 31 denotes a pattern memory in which sub-sequences (X ₀ , X ₁ , X ₂ ...) Are registered. In the figure, reference numeral 32 denotes an address control unit for reading unprocessed halftone image data held in the buffer memory 30 or writing new halftone image data. A reference numeral 33 corresponding to the first address control unit 4 in FIG. 1 is an address control unit for reading out a partial column held in the pattern memory 31 or writing a new partial column. A control unit 5, an exclusive OR circuit 34 corresponding to the exclusive OR circuit 3 in FIG. 1, a shift register 35, a mismatch counter 36, and a mismatch counter in FIG. counter 6, 37 is a threshold circuit corresponding to the threshold circuit 7 of FIG. 1, 38 is a maximum matching length detecting unit corresponding to the maximum matching length detecting unit 8 of FIG. 1, Reference numeral 39 denotes an encoding unit which creates a compression code as shown in FIG. 3 and converts the compression code into a Huffman code, for example.
Variable-length coding using a known method such as an arithmetic code and output terminal T ₂
This corresponds to the encoding unit 9 in FIG.

 Next, the operation of FIG. 5 will be described.

Halftone image data is input from an input terminal T _1, is accumulated in the buffer memory 30.

Now, as series x of the input symbol is the following, is x = aabababaa ......, when the component series _{x x = X 0 X 1 X} 2 ......, X 0 = λ ( empty string) and makes this first address The data is stored in the pattern memory 31 by the control unit 33.

First, when the first symbol a is read from the buffer memory 30 by the address control unit 32, the pattern memory
Since nothing is written in the column 31 except for the empty column, when the output of the pattern memory 31 by the address control unit 33 and the first symbol a are compared by the exclusive OR circuit 34, the exclusive OR circuit 34 The comparison result is output to the shift register 35 in parallel. When this is serially shifted and the number of “1” indicating the number of mismatches is counted by the mismatch number counter 26, the number becomes equal to or larger than the threshold Th, so that the threshold circuit 37 outputs a mismatch signal. As a result, the address control unit 33 stores this a in the pattern memory 31 as X ₁ = X ₀ a. Then, the index “zero” of this X ₀ a and the next symbol (the first “a”) are held in the maximum matching length detection unit 38,
These are applied to the encoding unit 39 to generate a compressed code as shown in FIG. 3, which is further compressed by a known method such as a Huffman code or an arithmetic code.

The address controller 32 reads the second symbol a from the buffer memory 30 and sends it to the exclusive OR circuit 34. At this time, the address control unit 33 stores the pattern memory 31
Read the X ₁ = X ₀ a from. Since X ₀ = λ and the empty column, the exclusive OR circuit 34 compares a and a at this time, and the comparison result is input to the shift register 35 in parallel, serially shorted to the mismatch counter 36, and The number of “1” in the serial output, that is, the number of mismatched bits is counted. When the second symbol matches, that is, the number of mismatches is within the threshold Th, the threshold circuit 37
The next symbol b and the registered sub-sequence are read out and compared by controlling the address control unit 32 and the address control unit 33, however, at this time, since no further sub-sequence is registered in the pattern memory 31, there is no match and the threshold circuit 37 outputs a mismatch signal. Thus the mismatch counter 36 is reset, and since the encoding unit 39 via the maximum matching length detection unit 38 components of the index and the next symbol is transmitted, X ₂ = X ₁ b and to words components index 1 and the next increment decomposition type codes created as shown in Figure 3 by the symbol b, which, while being stored in the pattern memory 31 as X _2, becomes compressed code, the output terminal is further compressed by the Huffman coding etc. It sent from T _2.

In this way, the pattern memory 31 creates and stores the incremental decomposition type code as needed.

When comparing the subsequence registered in the pattern memory 31 with the data in the buffer memory 30, the address control unit
Reference numeral 32 reads sequentially from the first word of the pattern memory 30, and the address control unit 33 sequentially reads each partial column registered from the pattern memory 31 from the first word of the partial column.

First, the first word of the pattern memory 30 is read by the address control unit 32 and output to the exclusive OR circuit 34. At this time, the address control unit 33 reads out each sub-sequence registered in the pattern memory 31 in order from the first word of the sub-sequence, compares and reads it with the exclusive OR circuit 34,
This result is set in the shift register 35 in parallel. Next, the shift register 35 is serially shifted, and the number of "1" s in the serial output is counted by the mismatch number counter 36, and thus the number of collation mismatch bits is counted. This count value is checked by the threshold circuit 37, and it is regarded as a match until the number of unmatched bits exceeds a predetermined threshold value or until the last data of the partial sequence in the pattern memory 31 is reached, and is regarded as one partial sequence. At this time, if the head word of the buffer memory 30 and the head word roughly match, the next word is read from the buffer memory 30 and compared with the next word of the roughly matched sub-sequence. Are the buffer memory 30 and the pattern memory
31 will be read out and compared. In this way, the comparison with each sub-sequence in the pattern memory 31 is sequentially performed.

The maximum matching length detection unit 38 holds an index indicating the order of registration of the obtained subsequence in the pattern memory 31, the length (the number of words), and the content of the next word (that is, the next symbol). Is detected, and the index and the contents of the next word are output. Encoding unit
39 is variable-length coded by the information transmitted from the maximum matching length detection unit 38 for example a Huffman code or the like, and outputs from the output terminal T ₂ as a compression code.

After the subsequence of the maximum matching length is encoded by the encoding unit 39 with the incremental decomposition type code, the portion extended by the encoded data of this subsequence is registered in the pattern memory 31,
Its encoded subsequence is discarded by the buffer memory 30, the same number of image data is input from the newly input terminal T ₁ to the buffer memory 30, is stored. Then, the same processing is performed from the next portion of the encoded subsequence, that is, a new head portion of the buffer memory 30.

Here, an example of the contents stored in the pattern memory 31 of FIG. 6 will be described with reference to FIG. FIG. 7 shows the contents stored in the pattern memory 31 by taking as an example the sequence described in the above (2) incremental decomposition type algorithm. The index in FIG. 7 (1) is the pattern memory.
31 addresses. The character string in FIG. 7 (2) is for reference only and is not actually stored in the pattern memory 31.

Each time a character string is decomposed into substrings, the pattern memory 31
(1) index, (3) last character, (4) 1
A pointer to a long character string, and (5) a pointer to the same character string except for the last character are sequentially registered. In this content, the item having no item (5) is the first character. The second and subsequent characters can be traced by item (4). Further, (5) it is possible to trace a character string that differs only in the last character by the same number of characters depending on the item.

In the above description, the incremental decomposition algorithm of the Jib-Lempel code is used, but the present invention is not limited to this, and another universal code may be used. For example, LZW code (TAWelch, "A Technique for High-Performance
Using "Data Comporession", Computer, June, 1984), the information of the compression code is only an index, which is simple.

〔The invention's effect〕

ADVANTAGE OF THE INVENTION According to this invention, the halftone images of various periods can be efficiently compressed without the mismatch of the halftone periods, and a small circuit scale can be realized without increasing the circuit for each halftone period as in the related art. it can.

In addition, a high compression ratio can be obtained without making image quality degradation noticeable by the information non-conservation type coding in consideration of the periodic regularity.

Of course, this restoration circuit can use the circuit using the universal coding system paired with the compression circuit without modification.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a diagram for explaining the principle of the present invention, FIG. 2 is a diagram for explaining the principle of a universal type Jib-Lempel code, FIG. 3 is a diagram for explaining an incrementally decomposed code, and FIG. 1 is a block diagram of an embodiment, FIG. 5 is a block diagram of a main part, FIG. 6 is a block diagram of a second embodiment of the present invention, FIG. 7 is an operation explanatory diagram, FIG. FIG. 9 is an explanatory diagram of a conventional predictive code restoration method. DESCRIPTION OF SYMBOLS 1 ... Buffer memory 2 ... Advance data memory 3 ... Exclusive OR circuit 4 ... First address control unit 5 ... Second address control unit 6 ... Number of mismatch counters 7 ... Threshold circuit 8 ... Maximum Matching length detection unit 9: coding unit

Claims (1)

(57) [Claims] An input data holding means (1) for holding image data to be compression-encoded, a preceding data holding means (2) for holding data subjected to compression coding processing, and a comparison means (3) ), A discrepancy number counting means (6) for counting the discrepancy number, and a threshold means (7) for detecting whether or not the discrepancy number is within a specified value.
And a maximum matching length detecting means (8) for detecting a maximum matching length; and an encoding means (9), and the input image data held by the input data holding means (1) is stored in the preceding data holding means (2). ) Is sequentially compared with the data held in the comparison means (3), and the maximum input data subsequence whose number of mismatches is within a specified value is collectively compressed and encoded by the encoding means (9). Dot image data compression method.