JPH0332280A - Dot image compressing system - Google Patents

Abstract

PURPOSE:To efficiently compress a dot image with the compact configuration of a hardware by adding a dissident run length to a maximum length sequence, registering it as the new sequence to a pattern accumulating means, compressing the code of this sequence and outputting it. CONSTITUTION:Partial columns in a pattern memory 4 are read and the almost coincident partial column with the maximum length is checked by a maximum length coincidence detecting part 11. The index (namely, registration order) of the partial column with the maximum length calculated in such a way in the pattern memory 4 and the dissident run length, which is corrected by an accumulating circuit error SIGMADELTALi, are outputted to an encoding part 12 and according to the combination of them, the compressed code is constituted and outputted. Thus, since compression and encoding can be executed to the similar run length pattern in a state as long as possible even when there is dissidence in a range set by a threshold value circuit 9 beforehand, dot image data can be efficiently compressed.

Description

[Detailed description of the invention] [Table of contents] Overview Industrial field of application Prior art (Figures 5 to 6) Problems to be solved by the invention Means for solving the problems (Figure 1) Working examples (Figures 2, 3, and 4) Effects of the invention [Summary] The present invention relates to a halftone image compression method, and aims to efficiently compress halftone images using a small hardware configuration. In a compression method for compressing value halftone image data, a run length creation means for creating a run length that is a continuous length of white pixels or black pixels, a storage means for temporarily storing this run length, and a run length creation means for creating a run length that is a continuous length of white pixels or black pixels; a pattern storage means for storing different series; a first calculation means for calculating a difference between a run length read from the storage means and a run length read from the pattern storage means; and two run length series. an accumulating means for calculating the sum of the differences between the first arithmetic means and the accumulating means; a threshold means for comparing the respective outputs of the first arithmetic means or the accumulating means with respective threshold values and outputting that either one exceeds the threshold; adjusting means for correcting the run-length series using the deviation information of the entire run-length series when the run-length series of the means is approximated by the run-length series of the pattern storage means; A sequence up to the end of the run-length sequence data length in the pattern storage means is regarded as one sequence, the maximum length of this sequence is detected, and information specifying the maximum length sequence in the pattern storage means for that sequence is outputted. and encoding means for encoding the maximum length sequence designation information of the storage means outputted from the maximum length coincidence detection means and the mismatch run length, add the new run length and register it as a new series in the pattern storage means,
It is also characterized in that it is compressed and encoded and output.

[Industrial application field]

The present invention relates to a halftone image compression method, and particularly to a compression method using a universal code.

In recent years, with the development of office automation, gradation images have come to be handled in addition to text as document information. Since documents are usually represented in binary values of white and black, in order to represent a gradation image, area modulation is used to determine the number of black pixels among halftone dots made up of multiple pixels according to the gradation. It is common to express using . When document information is used as digital data, the amount of data for a gradation image is much larger than that for a character image, ranging from several times to several ten times as much as the IO. Therefore, in order to efficiently handle gradation images in storage, transmission, etc., it is necessary to reduce the amount of data by applying efficient data compression.

[Conventional technology]

A modified read (MR) method is a conventional international standard compression method for binary images. 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 the change is such that there is little deviation of the black and white pattern boundary represented by the change point between adjacent scanning lines. This method compresses data by focusing on the connection relationships between points.

However, the gradation image formed by halftone dots cannot be effectively compressed using the MR method because the number of changing points caused by the halftone dots distributed over the entire screen is enormous.

For this reason, as a method for data compressing halftone images, a predictive coding method has been used in addition to the standard MR method.

In this method, as shown in FIG. 5, the pixel of interest d.

The reference pixel r (1
A method was used in which the black and white color of the pixel of interest dO is predicted by taking the data, and the prediction error is encoded. However, when compressing data by reading the notice with a scanner like in a facsimile, there are many types of halftone dots - the ζ halftone period is often unknown in advance. An adaptive predictive coding method was used in which a plurality of predictors using point periods as reference pixels are arranged, the predictor with the least number of mispredictions is selected, and coding is performed according to the selected predictor.

As shown in FIG. 6, the transmitted halftone image signal is input to the line memory 50, and its output is input to a first predictor 51 and a second predictor 52, each having a different halftone period, for prediction. In addition to determining the values, the exclusive OR circuits 53 and 54 detect whether these predicted values match the actual halftone image signal, and the number of mismatches is counted by the first prediction error counter 55 and the second prediction value. The prediction error is counted by a counter 56. The comparison circuit 57 compares the argument results for each fixed number of human dot image signals, and in the next section, the multiplexer 58 selects the predictor with fewer prediction errors, and the predictor with fewer prediction errors is selected. The prediction error signal was compressed and encoded by an encoder 59.

Therefore, if the prediction by the selected predictor matches the input halftone image signal, the prediction error signal output from the multiplexer 58 contains many "0"s and "1"s.
Since the number of bits is small, the compression efficiency of the encoder 59 is improved. This prior art adaptive predictive coding method is detailed in, for example, the Institute of Electronics and Communications Engineers Technical Research Report IE80-12r "Adaptive predictive coding of newspaper halftone photographs".

[The problem that the invention attempts to solve]

However, in the conventional data compression method for halftone dot images as described above, in order to configure a predictor by predicting the statistical properties of the halftone image, the halftone period of the prepared predictor is There has been a problem in that when the dot period matches the halftone period of the image, effective data compression is possible, but when it does not match, the efficiency of data compression deteriorates significantly.

By adopting such an adaptive predictive coding method, the problem of reduced data compression efficiency due to mismatched halftone dot periods can be improved to some extent, but if this improvement is to be made large, the predictor Since it is necessary to increase the number of circuits, there is a problem that the circuit scale increases.

Therefore, an object of the present invention is to provide a halftone image data compression method that can compress data while suitably matching the specific period with a small circuit scale in order to improve such problems. be.

[Means to solve the problem]

In order to achieve the above object, the compression method of the present invention, as shown in FIG.
The address control section 6, the first subtraction circuit 7, the accumulation circuit 8, the threshold circuit 9, the second subtraction circuit 10, the maximum length match detection section 11, the encoding section 12, etc. A universal encoder 2 is composed of a memory 4, a first address control section 5, a second address control section 1]6, and the like.

For the halftone image signal inputted from the input terminal T1, a run length creating section 1 creates alternate run lengths of white dots and black dots, and these run lengths are held in a buffer memory 3.

In the pattern memory 4, as will be described in detail in the embodiment section to be described later, when the incremental decomposition type Jibou-Lempel code is used, the existing component sequence is written.

As will be described in detail later, when using the incremental decomposition type Gibou-Lempel code, the encoding unit 12 outputs the index of the approximately matching encoded pattern component and the next symbol in the pattern memory 4 as a compressed code.

When the halftone image signal is transmitted from the input terminal T1, the run length creation section 1 creates a run length indicating the number of consecutive white dots and the number of black dots, and this is sent to the first address control section 5.
The data is written into the buffer memory 3 by the following.

Now the run length is white and black alternately -La,,Lb,L c
, , ia-. First address control unit 0 5 first subtract the run length La from the first subtraction circuit 7! Output this. Further, the second address control unit 6 sequentially reads out the written partial series from the pattern memory 4 from the beginning,
It is output to the first subtraction circuit 7. The threshold circuit 9 has a threshold Th for the output from the first subtraction circuit 7 and a threshold "rl+2" for the output from the accumulation circuit 8.

Therefore, when the second address control unit 6 outputs the head run length La of the first partial sequence from the pattern memory 4 to the first subtraction circuit 7 for the run length (-a),
When the difference 11, a-LΔ1, is greater than or equal to the threshold Th, a mismatch signal is output as a mismatch. As a result, the second address control unit 6 outputs the run length of the next partial sequence. In this way, if the run length La' read out from the pattern memory 4 is approximately the same as the above-mentioned La (when the difference is smaller than the threshold Thx), the threshold circuit 9 sends the match signal to the first memory controller 5. and output to the second memory control unit 6. As a result, the first address; νUl11 unit 5 outputs the next run length Lb from the hash sofa memory 3, and the second address control unit 6 outputs Lb' from the next part of the run length La from the pattern memory 4. do. At this time, the first subtraction circuit 6 calculates the calculation result (La-1-a
') is output to the accumulation circuit 8, and accumulation processing is performed. Therefore, at the beginning, this (La-1,, a') is held together with its positive and negative signs.

The comparison between Lb and Lb' read in this way is the first
It is performed in the subtraction circuit 7, and the calculation result (Lb-Lb')
is output to the accumulation circuit 9 and accumulated with the above (La-La'). This (La-La') + (Lb-Lb') is smaller than the above-mentioned interval (1! T h 2, and (Lb-Lb
') is smaller than the above-mentioned interval (iTh+), the threshold circuit 9 outputs a coincidence signal to the first address control section 5 and the second address control section 6, and similarly to the above, the next run length La is obtained from the hash sofa memory 3. , pattern me:[:S4 outputs the next partial sequence Lc' and compares it in the first subtraction circuit 7.As a result, when (Lc-Lc') is larger than the threshold Tht, the threshold (1) is determined as a mismatch by the circuit 9. The second subtraction circuit 10 outputs the Lc signal output from the buffer memory 3.
and the cumulative circuit error ΣΔL' = (1-aLa'
) The difference (Lc - ΣΔLi) from 10 (Lb - Lb') is taken as the mismatch run length, and this and the above-mentioned L a T and Lb
The index in the pattern memory 4 in which the subsequence of ' has been written and the matched run length La+Lb are held in the maximum length match detecting section 11. It should be noted that when the disagreement signal is output, the first subtraction circuit 7 and the accumulation circuit 8 are reset.

Then, the pattern memory 4 is read out again, and the Lc'
Read the run length L cII of the next subsequence of L
c-Lc'' and (La-La')→-(Lb-Lb'
) + (Lc-Lc'') are the thresholds Tht and T, respectively.
Check whether it is h2 or more.

In this way, the subsequences in the pattern memory 4 are read out, and the maximal length of the subsequences that roughly match is checked by the maximum length match detection section]1, and the pattern memory 4 of the subsequences with the maximum length thus obtained is The index (that is, the registration rank) within 9 and the mismatch length corrected by the cumulative circuit error ΣΔL1 are output to the encoding unit 12, and this combination constitutes a compressed code and outputs it.

[Effect]

Even if there is a discrepancy within the range preset by the threshold circuit, similar run length patterns can be compressed and encoded as long as possible, so halftone image data can be efficiently compressed.

That is, the human R1-sequence is approximated by the RL sequence with the longest approximate match in the pattern memory, that is, the RL sequence with the largest number of RLs within the allowable threshold. By doing so, as many RLs as possible can be approximately encoded within the tolerance.

〔Example〕

In the conventional halftone image data compression method, as explained with reference to FIGS. 5 and 6, the statistical properties of the image are predicted and a predictor is constructed from the predicted sample images. If the image is outside the expected range, the compression efficiency will be poor.

4 However, due to the periodicity of the halftone dots and the sameness of the halftone dot shapes, when multiple run lengths (the number of pixels between adjacent change points) are grouped together, there is a global regularity, so it is necessary to apply a universal code. By reducing the redundancy of this regularity,
This allows for effective compression.

Originally, universal codes are information-preserving data compression methods, and because they do not assume the statistical properties of the information source in advance when compressing data, they can be applied to various types of data (character codes, object codes, etc.). can.

In the present invention, in the process of compressing an input image, a universal code method is applied that optimizes the code while learning the statistical properties of the image, and performs efficient compression on various halftone images.

Before describing the present invention in detail based on one embodiment, a brief explanation of the universal code will be given (for details, see, for example, Munakata (Zjv-Lempel's Data Compression Method), Information Processing, Vol. 26, No. 1. 1985).Within the universal code there is an incremental 5 decomposition type (I incremental parsjn++).
) Ziv-Lempe1 code has a simple algorithm and is easy to calculate. Incremental decomposition type Z iv
-L In the empel code, if the sequence X of human symbols is x=aabababaa, then the incremental decomposition into the component sequence X=XOXIX2 is as follows.

If we rewrite the input symbol series X using the longest sequence of the existing components of the human symbols and the next symbol, we get the following.

x = a -a b -a b a-b-a a--
-----(ll) In this formula (1), the first raJ has no a in front of it, so a is added to all columns, and the next "a
Since raJ existed as an existing component in bJ, the next symbol rbJ was added to it, and the third l"aba" was "
Since abJ exists as an existing component, the next symbol r
aJ is added. In this way, if Xj is the longest string from which the rightmost symbol of the existing components is removed, the component series is expressed as follows.

Xo-λ (all columns), X3=X2a, X5=X+b.

X5=Xpa, X4=Xob, X5=Xs
It becomes a. For example, (in rabJ of formula 11, raJ is rX.

a'', so X5=X+b. Note that j in Xj represents a component index, and this index indicates the order in which each component appears.

Then, as shown in FIG. 3, encoding is performed using a set of the component index and the next symbol. Therefore, the above X3 is X3=X2
a, so it is encoded in the form of "2a". In this way, the incremental decomposition type algorithm searches for a maximum length match among the previously decomposed subsequences of the encoding pattern, and encodes it as a copy of the previously decomposed subsequence.

By the way, when such an encoding method is used, the compression rate improves as long subsequences match. The basic idea of the present invention is to detect roughly matching subsequences that are as long as possible without being concerned about slight differences when searching for matches between subsequences.

In other words, when an image that has already been halftone-dotted is read digitally, quantization noise is introduced for each pixel during reading, so even if the same gradation continues, it will not be accurately reproduced as the same halftone dot. do not have. Therefore, the importance of image quality at the single pixel level becomes less important, and gradation is expressed as pixel density, so it is not necessarily necessary to express it using information-preserving encoding, and it is better to do so as long as it does not affect image quality. It is important to increase the compression rate due to the non-information storage type.

In view of this point, the present invention incrementally decomposes a run length sequence, and when matching an existing component sequence with a sequence to be encoded, the difference in run length of each component is less than or equal to a predetermined threshold, and When the sum of the differences in each run length is less than a predetermined threshold, they are treated as the same sequence and treated as a match, and only when the degree of mismatch exceeds the threshold, it is considered a mismatch and decomposed into different subsequences. and the next symbol (
The difference in run length is adjusted using a value obtained by calculating the sum of the differences in the run length (run length).

An embodiment of the present invention will be described below with reference to FIGS. 2 and 3.

FIG. 2 is a block diagram of an embodiment of the present invention, FIG. 3(A) is an explanatory diagram of the encoding state, and FIG. 3(B) is an explanatory diagram of the operating state of the threshold circuit.

In FIG. 2, the same symbols as those in FIG. 1 indicate the same parts.

The run-length creation unit 1 alternately creates run lengths of white bisoto length and black bisoto length of the halftone image data that are manually inputted sequentially from the input terminal Tl, and registers 21 and 22.
, an exclusive OR circuit 23, an inverter 24, a run length counter 25, and the like. Register 21.2
2 and the 12-valued manual data is sequentially set and compared with the tJ1 transitive logical OR circuit 23. Therefore, “
When "1" or "-〇" is input continuously, the exclusive OR circuit 23 outputs "0" and the inverter 24 outputs "0".
1", the run-length counter 25 amplifies the output rlJ of the inverter 24 by a range of 1 to 1. However, when the white or black of the human image data comes to a point where it is interrupted, different binary data are stored in the two registers 21 and 22, and the logical sum circuit 23 outputs a reset signal "1". The run-length counter 25 sends the count value up to that point to the Hanofa memory 3 and starts the next counting process, so the count value up to that point is cleared.

In this way, run lengths indicating the continuous length of white or black in the halftone dot image are sequentially sent to the hash sofa memory 3.

The first address control section 5 performs control for writing the run lengths inscribed in the sofa memory 3 and output from the run length creation section 1, or for reading out the written run lengths.

Second a) The response control unit 6 performs control to write the run length outputted from the pattern memory 3 into the pattern memory 4, or to read out the written run length. That is, the second address control unit 6 has a function of adding the next data that does not match to the largest sequence and registering it as a new sequence in the storage means.

The first M arithmetic circuit 7 subtracts the run length output from the pattern memory 4 and the run length output from the pattern memory 4, and outputs the difference.

The accumulation circuit 8 accumulates the output of the first subtraction circuit 7.

The threshold circuit 9 includes a first threshold circuit a3i and a second threshold circuit 32, and the first threshold circuit 31 checks whether or not the output of the first subtraction circuit 7 exceeds the threshold (Ii!Th+). 2 threshold value circuit 32, the output of the accumulation port iS8 is the threshold value T112.
This is to check whether or not it exceeds the limit.

When a mismatch signal is output from the second subtraction circuit 10 and the threshold circuit 9, it is calculated based on the run length Lj that is determined to be a mismatch and the accumulated error value ΣΔL1 that is the output from the accumulation circuit 8 up to this run length LjO. (Lj − ΣΔLi), and calculate the run length ■−j by this (■, j−ΣΔLi
) and output to the maximum length matching unit 11. Through such replacement processing, the entire length of image data can be adjusted.

1 Next, the reason for this will be briefly explained with reference to FIG. 3(B).

The run length output from Hasofa memory 3 is La,
Lb, jc-, and when comparing them with the run lengths La', Lb', LC'- output from the pattern memory 4, an example will be described in which the threshold values T h l = 4 and Thx = 5, for example.

First, the first subtraction circuit 7 calculates the difference between La' and Ta, resulting in the result La -La-3-----(A), and then calculates the difference between b' and Lb. Result L b
' −L b = 2−−−−−− (B),
Furthermore, the result of calculating the difference between Lc' and Lc is L c ' - L
Consider the case of c = 6 (C).

Therefore, Δ1 in FIG. 3(B) is 3 obtained from the above formula (A), and becomes Δ1-3. At this time, since 3 is smaller than Th+, the threshold circuit 9
Therefore, no mismatch signal is output. Subsequently, the calculation of equation (B) is performed, so Δ2 in Fig. 3 (B) becomes (A) + (B), and Δ2 = -3+
2---1. At this time, since the (-river) of 2 and Δ2 in equation (B) are smaller than the Th+ and Thp, respectively, no mismatch signal is output from the threshold circuit 9. Following this, operation (C) 2 is performed, but since 6 in equation (C) exceeds Th+, Δ3-Δ2 + 6 = -14-6= 5,
There is no need to detect that this Δ3 is equal to or greater than T h 2, and the threshold circuit 31 outputs a mismatch signal, thereby causing the threshold circuit 9 to output a mismatch signal. As a result, when the second subtraction circuit 1o outputs the mismatched run length series from the panzer memory 3, the Δ2−
-1 is subtracted from LC(-1,)=Lc4-1=Lc'', which is replaced with Lc and sent to the maximum length -Pi detection unit 11.

This makes it possible to obtain La'+Lb'+Lc - La+Lb+Lc.

Next, the operation of the embodiment of the present invention shown in FIG. 2 will be described.

(1) Binarized halftone image signals are inputted sequentially from the input terminal T1 and are sequentially stored in two registers 21 and 22 of one register each connected in series. The exclusive OR circuit 23 connects these two registers 21.2
2. When the binary data stored in these two registers 21 and 22 are both "1" (black) or "0" (white), the exclusive OR circuit 23 outputs "1" from the inverter 24. , run length counter 2
Add a count amplifier signal to 5. In this way, the run-length counter 25 counts the length (run length) in which "1" or "0" continues in the image data.

Then, at the point where "1" or "0" ends, different binary data are stored in the two registers 21 and 22, so the exclusive OR circuit 23 outputs "1" and the run length counter 25 A recent signal "1" is output to the

3 Upon receiving this reset I-signal "1", the run-length counter 25 transmits the count value up to that point to the panzer memory 3 of the universal encoder 2, and in order to start the next counting process, the run length counter 25 transfers the count value up to that point to the panzer memory 3 of the universal encoder 2. Clear. In this way, run lengths of 1-1'' or 0 in the halftone image are sequentially stored in the panzer memory 3.

(2) When the run length series of "0" or rlJ, LB, LC, LD=LE is currently stored in the panzer memory 3, the first address control unit 5 first reads out the run length series. The second address control unit 6 reads out the first run length of the registered run length series from the pattern memory 4 and sends it to the first subtraction circuit 7. and subtract. If this subtraction value is larger than the threshold Th+, it is determined that there is no match, and the process moves on to the next run-length series comparison. In this way, when LA is matched with all run length sequences in the pattern memory 4 and still does not match, the second
The address control section 6 sends an index indicating the registration order to the maximum length match detection section 54 and the peak section 11, and the second subtraction circuit 10 sends the run length of LA to the maximum length match detection section 11. The maximum length coincidence detection section 11 transmits these to the encoding section 12, so the encoding section 12 performs variable length encoding on this, for example, and outputs it as a compressed code.

After the run length LA is encoded by the encoder I2 in this way, the second address control unit 6 registers it in the pattern memory 4 as a new run length sequence, and the first address control unit 5 This LA is discarded from the front memory 3 and a new next run length is written in the front memory 3.

(3) Next, the first address control unit 5 controls the Panzer memory 3
The run length series is then read out and sent to the first subtraction circuit 7. Further, the second address control unit 6 also sequentially reads out the registered head run lengths from the pattern memory 4 in the same manner as described above. This roughly matches the leading run length sequence' (when the difference is within the threshold (i!Th1, Th2))
In this case, LB'-LB-ΔLB is ΔL≦T
If it exceeds hI, ΔL6, it matches up to the first RL in the pattern memory. From then on, comparison with the longer R1 series is no longer necessary.

When configuring a pattern memory as shown in Appendix 1, the acid index does not need to be registered as it becomes the address of the memory. Apart from this configuration, the last RL of the RL series
It is also possible to adopt a configuration in which the RL sequences and indexes are registered in the pattern memory. In this case, the positions where the RL sequences are registered in the pattern memory do not need to be arranged in order.

In addition, in the above explanation, ZivLemp
el) Although the code incremental decomposition algorithm is used, the present invention is of course not limited to this, and other universal code systems may be used. For example, LZW code (T, A
Welch, “A Technique for
HighPerformance Data Comp
”Computer, June, 198
4), the compression code is only affected by the index, making it simpler.

1 [Effects of the Invention] According to the present invention, halftone dot images of various periods can be efficiently compressed without mismatching the halftone dot periods, and unlike conventional methods, the number of circuits does not increase for each halftone period. It can be realized with a small circuit scale. Also, by using non-information-preserving encoding that takes periodicity into account, a high compression rate can be obtained without noticeable deterioration in image quality.

[Brief explanation of drawings]

Fig. 1 is a diagram of the principle of the present invention, Fig. 2 is a configuration diagram of an embodiment of the present invention, Fig. 3 (A) is a diagram explaining the code state, (B) is a diagram explaining the operating state of the threshold circuit, Fig. 4 5 is an explanatory diagram of a pattern memory, FIG. 5 is an explanatory diagram of a conventional predictive encoding method, and FIG. 6 is an explanatory diagram of a conventional predictive code restoration method. 1-Run length creation section 2-Universal encoder 3-Hasophor memory 2 Pattern memory First address control section Second address control section First subtraction circuit Threshold circuit Maximum length match detection section Encoding section 8-Accumulation circuit 1 ( L--1 second subtraction circuit

Claims (1)

[Claims] A compression method for compressing binary halftone image data, comprising: run length creation means (1) for creating a run length that is a continuous length of white pixels or black pixels; a storage means (3) for temporarily storing; a pattern storage means (4) for storing different series of run lengths; and a run length read from the storage means (3);
a first calculating means (7) for calculating the difference between the run lengths read from the pattern storage means (4); an accumulating means (8) for calculating the sum of the differences between the two run length series; 1 arithmetic means (7) or the accumulation means (8), each output is compared with a respective threshold value, and a threshold value means (10) outputs that either one has been exceeded; and the storage means (3). adjustment means (10) for correcting the run-length series using deviation information of the entire run-length series when the run-length series is approximated by the run-length series of the pattern storage means; The length of the run-length sequence data in the pattern storage means (4) is defined as one sequence, and the maximum length of this sequence is detected, and the length of the run-length sequence data in the pattern storage means (4) is a maximum length match detection means (11) for outputting information specifying a maximum length sequence of the maximum length sequence of the storage means (3) outputted from the maximum length match detection means (11); The apparatus is equipped with an encoding means (24) for encoding the maximum length sequence, adds the run length that has become inconsistent to this maximum length sequence, registers it as a new sequence in the pattern storage means (4), and compresses and encodes this as an output. A halftone image compression method characterized by: