JPH06311372A - Data compression method and device for binary picture - Google Patents

Abstract

PURPOSE:To attain adaptive coding in response to property of a picture by generating reference data of Markov model coding in a form including the result of area discrimination in response to a structural statistic quantity from a binary picture. CONSTITUTION:An area discrimination section 11 discriminates at which area a j-th noticed picture element Xi,j of a 1st line of an input picture is in existence. Picture element period data from a period arithmetic operation section 16 and an estimate density L*i,j from a change point Vi,j estimate density arithmetic operation section 15 calculated by a dispersion arithmetic operation section 14 are used. The discrimination content indicates whether an object picture element is a flat part of a picture or a contour, a part with a period of picture element or without the period. Whether or not the picture is a part with a period of picture element is immediately discriminated by an output from a period arithmetic operation section 16, then the discrimination is a major operation of the discrimination section 11. Thus, the adaptive Markov model coding is performed in response to the property of the picture, then adaptive coding in response to the property of the picture is executed.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a binary image coding method and apparatus, and more particularly, to a document drawing, line drawing, halftone dot drawing,
The present invention relates to a data compression method and device for a binary image in which photographs and the like are mixed.

[0002]

2. Description of the Related Art As a standard method for encoding a binary image in which each pixel is represented by "0" or "1", MH (Mode fied Huffman) encoding or MR (Mod
efiedREAD, Modefied Relative Element Address Desig
nate) encoding, which is CCIT as an international standard.
Established by T (International Telegraph and Telephone Advisory Committee). Although these MH coding and MR coding are suitable for facsimile transmission of office documents, etc., binary images (mixed binary images) in which text, line graphics, halftone dots, photographs and the like are mixed are provided with high definition. It is insufficient for transmission, and is not suitable for application to display an image on a user terminal in image database search and the like.

Therefore, targeting binary still image data,
Encoding with a high data compression rate is possible even when text and line graphics are mixed with halftone dot drawings and photographs, and hierarchical display is possible on the user terminal side. As a method, JBIG (Joint Bi-level
The Image coding experts Group) method has been proposed as an international standard. A binary image data compression apparatus based on the JBIG encoding method is roughly configured by a portion that is modeled by a binary Markov model that refers to adjacent 10 pixels and an encoder that performs entropy encoding. As entropy coding, arithmetic coding is adopted, and a QM code developed based on Q code and Mel code (hereinafter referred to as QM-Code).
Has been decided to be used as the standard code. The JBIG method will be further described below with reference to FIGS. 11 and 12.

FIG. 11 shows a conventional binary image data compression apparatus based on the JBIG method. Here, it is assumed that the input binary image is scanned by raster scanning in which pixels are sequentially scanned from the upper left of the image.
In the figure, x _{i, j} represents the j-th pixel (target pixel) input on the i-th line. This apparatus includes a correlation calculation unit 91 that calculates a correlation of images, a pixel movement position determination unit 92 that changes the position of a pixel having a strong correlation, and a reference data creation unit 93 that generates reference data in Markov model coding. And the QM coder 9 that actually encodes
4 and the output of the QM coder 94 becomes the encoded output. In this device, the pixel of interest x _{i, j} is Q
When encoded by the M coder 94, the Markov model encoding is performed based on the reference data output from the reference data creation unit 93. In addition, as the reference data of the Markov model encoding, the pixel of interest x to be encoded is
Ten pixels adjacent to _{i, j} (ten adjacent pixels) are used. FIG. 12 shows the positional relationship between the pixel of interest x _{i, j} and its 10 adjacent pixels. Further, in this apparatus, it is considered that the reference pixel position of the reference data can be changed by one pixel according to the correlation of the image. The changing method is based on the data calculated by the correlation calculating unit 91 using the peripheral pixels, by the pixel moving position determining unit 92,
The pixel at the position A shown in FIG. 12 among the 10 adjacent pixels is to be replaced with a pixel (replacement pixel) at another pixel position having a strong correlation. That is, the replacement data is used instead of the pixel A to create the reference data. However, the information on the position of the replacement pixel needs to be transmitted from the encoding side to the decoding side.

[0005]

By the way, QM-co
Conventional JBI for Markov model coding using de
The G method has problems to be solved as described below.

(1) Since the adjacent 10 pixels are used as the reference data for the Markov model coding, the compression efficiency of a binary image such as a document drawing or a line drawing is high, but a pseudo halftone image using the error diffusion method is used. Compression efficiency is low.

(2) It is possible to replace only one pixel out of the ten adjacent pixels of the reference data with the pixel at the pixel position having a strong correlation. However, with this alone, the pseudo image using the halftone image and the tissue dither method is used. It is not sufficiently effective for an image in which binary image data such as a halftone image has a specific period. That is, in the binary image data having these specific periods, a plurality of pixels that are apart from _{the target} pixel x _{i, j} by a specific period have a strong correlation with the target pixel, and it is not possible to change the pixel position of only one pixel. It is enough.

(3) In changing the pixel position of the reference data,
The pixel position can be moved only in line units of the image. However, when document texts and pseudo-halftone images coexist, it is general that document texts and pseudo-halftone images coexist in one line. In such a case, the pixel position of the reference data cannot be changed. .

(4) When the pixel position is changed, it is necessary to calculate the correlation with surrounding pixels and the Markov model entropy in order to obtain the moving position of the pixel. Since these arithmetic processes require a large amount of calculation and also require a large-capacity memory for temporarily storing data, they impose a heavy cost or system burden on implementation. .

It is an object of the present invention to provide a method and apparatus capable of efficiently performing data compression on general binary image data including mixed images regardless of the type of binary image data.

[0011]

SUMMARY OF THE INVENTION A binary image data compression method according to the present invention is a first method for obtaining image statistics from an input binary image.
, A second step of performing area determination of the input binary image according to the image statistics and generating area determination data, and a third step of generating reference data in Markov model encoding. However, the reference data is composed of a first data portion corresponding to the area determination data and a second data portion which is a data portion other than the first data portion.

A binary image data compression apparatus of the present invention performs image statistic calculation means for calculating an image statistic from an input binary image, and performs area determination of the input binary image according to the image statistic. Area determination means for generating area determination data, and reference data generation means for generating reference data in Markov model coding, wherein the reference data corresponds to the area determination data and the first data portion and the first
And a second data portion that is a data portion other than the data portion of.

[0013]

According to the binary image data compression method of the present invention, the structural statistical amount of the image is obtained from the input binary image, the region of the input binary image is determined according to this statistical amount, and the Markov model coding is performed. Since the reference data in 1 is created in a form that includes the result of this area determination, it is possible to perform adaptive Markov model coding according to the property of the image. As a result, a large amount of data compression is realized as compared with the conventional case in which only encoded adjacent pixels are used as reference data for encoding. It is also possible to divide the input binary image into appropriate blocks for region determination, but considering compatibility with Markov model encoding, which is non-block encoding, the region determination is performed for each pixel of the input binary image. It is preferable to carry out every time.

The binary image data compression method of the present invention creates reference data used for Markov model coding. Following the creation of reference data, an input binary image is created using this reference data. By providing the fourth step of performing the Markov model encoding, the adaptive reference data is used for encoding, and it is possible to encode the mixed image at a high data compression rate. Becomes When area determination is performed for each pixel,
In the fourth step, it is desirable to perform the Marco model coding of each pixel by distinguishing the Markov state of each pixel region using the above-mentioned reference data, and by doing so, at a higher compression rate. Data encoding is achieved.

In the binary image data compression method of the present invention, the reference data is composed of a first data portion corresponding to the area determination data and a second data portion which is a data portion other than the first data portion. Has been done. It is desirable that this second data portion also be adaptively determined according to the area determination data. Specifically, the second data portion can be selected from a plurality of data that should be candidates for the second data portion according to the area determination data. As an example of such a second data portion, the second pixel is composed only of the pixel values of the coded pixels adjacent to the pixel of interest, the estimated density value of the pixel of interest calculated from the coded pixels, A combination of one encoded pixel value having a strong correlation with the pixel of interest and the pixel value of a pixel adjacent to the pixel of interest, and further,
A combination of the estimated density value of the target pixel calculated from the coded pixels, a plurality of coded pixel values that have a strong correlation with the target pixel, and the pixel values of pixels that are close to the target pixel , And so on. Second
How to configure the data portion of is determined depending on what kind of image is mainly treated as an input binary image, a desired data compression rate, the required amount of hardware or software, and the like.

In the binary image data compression method of the present invention, what is taken as a criterion for area determination can be appropriately determined. However, in order to make the reference data more adaptive in consideration of the properties of Markov model coding, whether the pixel of interest is in the contour part or the flat part of the image, the period in the input binary image is It is preferable to determine whether or not there is a structure.

In the binary image data compression apparatus of the present invention,
An image statistic calculating means for obtaining a structural statistic of an image from an input binary image, an area determining means for determining an area of the input binary image according to the statistic, and reference data in Markov model coding Since it has a reference data generation means that is created in a form that includes the result of region determination, it is possible to perform adaptive Markov model coding according to the nature of the image. As a result, a large amount of data compression is realized as compared with the case of using a conventional data compression device that encodes only adjacent pixels that have already been encoded as reference data. It is also possible to divide the input binary image into appropriate blocks for region determination, but considering compatibility with Markov model encoding, which is non-block encoding, the region determination is performed for each pixel of the input binary image. It is preferable to carry out every time.

The binary image data compression apparatus of the present invention creates reference data used for Markov model coding. Further, the Markov model coding of the input binary image is performed using this reference data. By providing the encoding means for performing, the encoding using the adaptive reference data is performed, and the encoding of the mixed image with a high data compression rate can be performed consistently from the image input. Becomes When the region determination is performed for each pixel, it is desirable that the encoding unit uses the above-mentioned reference data to distinguish the Markov state of each pixel region and perform the Marco model encoding of each pixel. By doing so, data encoding with a higher compression rate is achieved.

[0019]

Embodiments of the present invention will now be described with reference to the drawings. FIG. 1 is a block diagram showing the configuration of a binary image data compression apparatus according to an embodiment of the present invention.

This binary image data compression apparatus includes an image statistic calculator 10 for calculating an image statistic from input binary image data, and a binary value input for each pixel based on the calculated image statistic. Area determination unit 11 for determining an area of an image
, A reference data creation unit 12 that generates reference data in Markov model coding, and a QM coder 13 that performs Markov model coding by referring to the reference data. The image statistic calculator 10 is an image statistic calculator, the region determiner 11 is a region determiner, the reference data generator 12 is a reference data generator, and the QM coder 13 is an encoder. The image statistic calculator 10 is provided with a variance calculator 14, an estimated density value calculator 15, and a cycle calculator 16. The output of the QM coder 13 is the encoded output of this device. When the reference data creation unit 12 generates reference data, the area determination unit 11
The area determination data corresponding to the area determination result in (1) is added to the reference data. The binary image scanning method input to this apparatus is a raster scanning method in which sequential scanning is performed from the upper left portion of the image. x _{i, j} is
It represents the j-th pixel in the i-th line of the image.

When the pixel of interest x _{i, j} is coded, the variance calculating unit 14 calculates the change points of the 10 adjacent pixels that have already been coded, and outputs the number V _{i, j} of the change points. Is. Specifically, for the target pixel x _{i, j} , 10 adjacent pixels surrounded by the thick solid line in FIG. 2 are targeted, and the pixel values are different on both sides of 12 boundaries between these adjacent pixels. It is determined whether there is a difference, and the number of boundaries having different pixel values is output as the number of change points V _{i, j} . The boundaries between adjacent pixels are indicated by thick broken lines in the figure.

FIG. 3 shows an example of the configuration of the distributed arithmetic unit 14. The variance calculating unit 14 accumulates the input image data for three lines to determine the proximity 10 to the pixel of interest x _{i, j} .
A line memory 21 for storing pixels (pixels indicated by diagonal lines in the figure), 12 exclusive OR circuits 22 _{1 to} 22 ₁₂ respectively corresponding to boundaries between adjacent pixels, and exclusive OR
It is configured by an adder 23 that counts the number of the outputs of the circuits 22 _{1 to} 22 ₁₂ that are "1". The values of pixels located on both sides of the corresponding boundary are input to each exclusive OR circuit. For example, the values of the pixels x _{i-2, j -1} and x _{i-2, j} are input to the _first exclusive OR circuit 22 ₁ . If the values of these pixels x _{i-2, j-1} and x _{i-2, j} are different from each other, exclusive OR
The output of the circuit 22 ₁ is "1", and if they are the same, the output is "1".
0 ". Similarly, the other exclusive OR circuits 22 _{2 to} 22 22
The output of ₁₂ also becomes "1" when the pixel values are different on both sides of the corresponding boundary. Therefore, by counting the number of exclusive OR circuits having "1" by the adder 23, the number of change points ( The variance value V _{i, j} will be obtained.

The estimated density value calculator 15 determines the pixel of interest x _{i, j}
The pixel of interest x _{i, j} is estimated from the values of the coded pixels which are the pixels around the pixel and is output as the estimated density value L ^* _{i, j} . An example of a calculation formula for _{obtaining the} estimated density value L ^* _{i, j} is shown below. The estimated concentration value L ^* _{i, j} is, for example, 5
Represented as a bit value.

[0024]

[Equation 1] Here, x _{im, jn} is a coded peripheral pixel that does not include the _target pixel x _{i, j} , and its pixel value is “0” or “1”. FIG. 4 shows an example of an arrangement of peripheral pixels used when _obtaining L _{i, j in the} equation (1). 3 within the thick solid line
Six pixels are used to calculate L _{i, j} .

FIG. 5 is a block diagram showing an example of the structure of the estimated density value calculator 15. The estimated density value calculation unit 15 stores the input image data for five lines to store the above-mentioned 36 peripheral pixels (pixels indicated by diagonal lines in the figure) for the pixel of interest x _{i, j} , and these line memories 31. An adder 32 that adds pixel values of 36 peripheral pixels to calculate L _{i, j} according to the equation (1), a line memory 33 that accumulates the calculated value of L _{i, j} , and an equation (2) And an arithmetic unit 34 for calculating the estimated density value L ^* _{i, j} in accordance with the above.

The period calculator 16 obtains the periodicity of the pixel from the encoded peripheral pixels and outputs the obtained pixel period data. FIG. 6 is a block diagram showing an example of the configuration of the period calculator 16. An 8-stage shift register 41 for sequentially inputting the value of the pixel of interest x _{i, j to} the least significant bit (LSB) is provided.
The 2-bit output on the most significant bit (MSB) side of the first is the first for calculating the convolution sequence P2 _{i, j} of two pixels in the period.
Connected to the exclusive OR circuit 42 ₁ . The output of the next 2-bit counted from the most significant side of the shift register and the output of the first exclusive OR circuit 42 ₁ are the second exclusive for calculating the convolutional sequence P4 _{i, j} of period 4 pixels. Input to the logical OR circuit 42 ₂ . Then, the output of the least significant 4 bits of the shift register and the output of the second exclusive OR circuit 42 ₂ are the third exclusive OR circuit for calculating the convolution series P8 _{i, j} of the period of 8 pixels. It is connected to 42 _3.

Convolution P4 _{i, j of} period 4 pixels and period 8
The convolution series P8 _{i, j} of pixels are input to the shift registers 43 and 44, respectively.
On the output side of 44, each convolutional sequence P4 _{i, j} , P8 _{i, j}
Change point counter 4 for counting the number of change points in
5,46 are provided. That is, the change point counter 4
5 calculates the number of change points in the bit string from P4 _{i, j} to P4 _{i, j-7} . These change point counters 45, 46
The outputs of are each 3 bits wide. Then, for each of the change point counters 45 and 46, AND circuits 47 and 48 to which the 3-bit output from the change point counter is input are provided, and an AND circuit 49 to which the outputs of these AND circuits 47 and 48 are input is provided. It is provided. Last AND
The output P _{i, j} of the circuit 49 indicates whether or not the pixel has a cycle. When P _{i, j} is “0”, it indicates that the pixel has a cycle, and when it is “1”. This indicates that the pixel has no cycle. Further, the output of the AND circuit on the side of the 8 pixel period is P4.
/ 8 signal. The value of the P4 / 8 signal indicates the period of the pixel, and "0" indicates the 8-pixel period, and "1" indicates the 4-pixel period. After all, if there is no change point between consecutive 4 pixels or 8 pixels in the convolutional series P4 _{i, j} , P8 _{i, j} , the period calculating unit 16 respectively determines the pixel corresponding to the pixel x _{i, j.} It will be determined that the sequence has a period of 4 pixels or 8 pixels.

Next, the area determination unit 11 will be described.
The area determination unit 11 determines, for the j-th target pixel x _{i, j of the i-} th line of the input image, what area the pixel x _{i, j} is in. For this determination, the number of change points V _{i, j} calculated by the variance calculation unit 14, the estimated density value L ^* _{i, j} calculated by the estimated density value calculation unit 15, and the pixel cycle calculated by the cycle calculation unit 16 The data is used. The contents of the determination are whether the target pixel is in the flat portion or the contour portion in the image, and whether it is in the portion having the pixel period or in the portion having no pixel period. . Here, since it is immediately known from the output from the period calculation unit 16 whether or not it is in the portion where the pixel period is present, the region determination unit 11 mainly determines whether it is in the flat portion or the contour portion. It will work.

FIG. 7 is a block diagram showing an example of the configuration of the main part of the area determination section 11. The area determination unit 11 includes
A line memory 51 that stores the estimated density value L ^* _{i, j} input from the estimated density value calculation unit 15, and a maximum value L ^* _max and a minimum of the estimated density value L ^* _{i, j} accumulated in the line memory 51. A maximum value detection unit 52 and a minimum value detection unit 53 that detect the value L ^* _min are provided. Also, the maximum value L ^* _max
And a minimum value L ^* _min , the subtracter 54 calculates the difference, the first comparator 55 that compares the output of the subtractor 54 with the first threshold value Th _1, and the change from the variance calculator 14 A second comparator 56 is provided which compares the score V _{i, j} with the second threshold Th ₂ . The output of the first comparator 55 is L ^* _max
-L ^* _min> "1" next to the case of Th _1, becomes "0" in other cases. The output of the second comparator 56 is V _{i, j} > T
In case of h ₂ , it becomes “0”, and in other cases, it becomes “1”. Further, there is provided a concentration area separation unit 57 which determines whether or not the estimated concentration value L ^* _{i, j} is between the third and fourth threshold values Th ₃ and Th ₄ . The output of the density region separation unit _57, Th ₃ <L ^* _i, becomes "1" when the _j ≦ Th _4, the otherwise becomes "0". The outputs of the second comparator 56 and the density region separation unit 57 are input to the AND circuit 58,
The outputs of the first comparator 55 and the AND circuit 58 are input to the OR circuit 59. The output A _i of the OR circuit _{59, j} is a which represents whether the contour is a planar portion, A _{i, j}
"0" indicates that the target pixel exists in the flat portion, and "1" indicates that the target pixel exists in the contour portion.

Next, the procedure of area determination by the area determination unit 11 will be described. In the device of this embodiment, the pixel of interest x
_{When i, j} is encoded by the QM coder 13, the area determination unit 11 first determines the area of the target pixel x _{i, j} . The area determination unit 11 to which the number of change points V _{i, j} , the estimated density value L ^* _{i, j} , and the pixel cycle data are input, performs area determination on a pixel-by-pixel basis. First, the estimated density value L ^* _{i, j of the target} pixel x _{i, j} and the estimated density value L ^* _{im, jn} (m = 0,1, ..., N = 0,1, ..., but m
² + n ² ≠ 0), the maximum value L ^* _max and the minimum value L ^* _min are _calculated, and the difference between the maximum value L ^* _max and the minimum value L ^* _min is calculated. Then, for the difference, the pixel area is determined according to the following determination conditions. The output of the first comparator 55 corresponds to this determination result.

[0031]

[Equation 2] Pixels determined to be flat part candidates under this determination condition are
Pixels in the density flat part of the pseudo-halftone image binarized using the error diffusion method or the tissue dither method, or the high resolution fine halftone dots, the character background, the thick line, and the area in the middle of the character, etc. . The pixels determined to be the contour portion are pixels in areas such as characters, line graphics, coarse halftone dots with low resolution, and the contour portion of the pseudo halftone image.

By the way, as the characteristic of the density flat portion of the pseudo-halftone image, in addition to the characteristic that the change amount (L ^* _max- L ^* _min ) of the estimated density value becomes small as in the above-mentioned determination condition, white It has the feature that the number of change points with black becomes large. This is because when a flat portion of a halftone image such as a photograph is binarized, white and black pixels are evenly distributed, and many white or black isolated points are generated. Treating an area where the number of change points between white and black is large as a flat portion hinders subsequent encoding. Therefore, such an area is deleted from the area determined to be a flat portion and treated as a contour portion. There is a need. Therefore,
In order to delete the texture area of the binary image from the target pixel x _{i, j} which is determined to be the flat part candidate, the proximity 10
The area determination is corrected using the number of change points of the pixel. Specifically, if it is determined that the flat part candidates, on which had been set a threshold Th ₂ to TH ₄ from previously second to fourth, obtained by the variance calculating section 14 changes Points V _{i, j} ,
Using the estimated density value L ^* _{i, j} , the flat portion and the contour portion are separated according to the following determination conditions.

[0033]

[Equation 3] In this way, it is determined whether the pixel x _{i, j} belongs to the contour portion or the flat portion, and the determination result A
_{i, j} will be output.

Subsequently, the area determination section 11 is provided with a cycle calculation section 16
On the basis of the pixel cycle data from 1), the contour portion and the flat portion are distinguished by the presence or absence of the pixel period. As a result, the binary image is divided into the following four types of regions I to IV on a pixel-by-pixel basis.

(I) Pixel periodicity and flat part → texture dither image, fine halftone dots: (II) Pixel periodless and flat part → error diffusion image, background, thick lines and characters: (III) Pixel Periodic and outline part → Coarse halftone dots: (IV) No pixel period and outline part → Character, line figure, pseudo halftone image outline part:

In this way, in the area judging section 11,
The area determination is performed for each pixel, and the result of the area determination is sent to the reference data creating unit 12 as the determination data.

Next, the reference data creating section 12 will be described.
To do. The reference data creation unit 12 uses the Markov model encoding
To generate reference data in
Whether the input pixel signal and area determination unit 11 is included in the creation unit 12
In addition to the judgment data from
Signal P4 / 8 indicating the
Estimated concentration value L from^* _{i, j}Is typing. Generated reference
An example of the data structure is shown in FIG. Illustrated
The reference data is 12 bits wide, and the area determination data 2
Bit, pixel of interest x_{i, j}Pixel data of 6 pixels close to each other 6
Bits and adaptive data 4 bits.
Each bit of the area determination data is a flat part / contour
Data that indicates the part and data that indicates the presence or absence of a pixel cycle.
is there. The 6 adjacent pixels are the pixels represented by the thick solid line in FIG.
(X_{i-2, j}, x_{i-1, j-1}, x_{i-1, j}, x _{i-1, j + 1}, x_{i, j-2}, x
_{i, j-1}), Regardless of the result of the area determination.
Is included in the data. Below, this 6-pixel proximity
And will be referred to as fixed reference pixel data. Meanwhile, adaptation
The content of the data changes depending on the area determination result.
The data. That is, the area determination data is I to
Depending on the case of IV, the data will be as follows.
It

[0038]

[Table 1] Here, the four adjacent pixels other than the fixed reference pixels are, for example, pixels x _{i-1, j-2} , x _{i-2, j -1} , x _{i-2, j + 1} , x _{i-1, j +. Is 2} . The periodic pixel is a pixel having a strong periodic correlation with the pixel of interest x _{i, j} when the periodicity of the image from the encoded peripheral pixels is analyzed. k is an integer of 0 to 4, and the number of pixels used as the periodic pixels changes according to the strength of the periodic correlation. Furthermore, the estimated concentration value L
^* The quantized data of _{i, j} is the state-compressed information of the coded peripheral pixels.

Next, the specific structure of the reference data creation unit 12 will be described.
An example will be described with reference to FIG. This reference day
The data creating unit 12 corresponds to the case where k is 0 to 1.
The area determination data (2 bits
,), Fixed reference pixel data (6 bits), and density estimation
Value L^* _{i, j}(5-bit width), signal P4 / 8, fixed reference
Proximity pixel data other than pixel (x_{i-2, j-1}, x_{i-2, j + 1}, x
_{i-1, j-2}, x_{i-1, j + 2}, x_{i, j- Four}, x_{i, j-8}) And enter
It Of these, the area determination data and the fixed reference pixel data are
The reference data creation unit 12 outputs the data as it is.
Then, the reference data creation unit 12 displays the estimated density value L^* _{i, j}
First and second quantizers 61 and 6 for further quantizing
2. Pixel x according to the signal P4 / 8_{i, j-4}And x_{i, j-8}Nozu
Selector 63 for selecting one of
Adaptive data selection unit 64 for generating adaptive data according to a value
It is composed of and. First and second quantizer
61 and 62 are estimated density values L, respectively.^* _{i, j}3 bits
Width value L^** _{i, j}And a 4-bit wide value L^*** _{i, j}Convert to
It is something. In the selector 63, the signal P4 / 8 is "0".
If pixel x_{i, j-8}Is selected, and if "1", pixel x_{i, j-4}
Is configured to select. On the other hand, adaptive data selection
The selecting unit 64 is P_{i, j}And A_{i, j}Depending on the value of
It is configured to select an input signal. Where x
_{i, j-4}/ X_{i, j-8}Is x _{i, j-4}And x_{i, j-8}Selector from
The one selected by 63 is shown.

[0040]

[Table 2] By configuring the reference data creation unit 12 in this way, the region determination data according to the region determination result in the region determination unit 11 is added to the reference data, and the adjacent pixels are adaptively adjusted according to the region determination data. Data or data representing the state of peripheral pixels is added. The reference data output from the reference data creation unit 12 is input to the QM coder 13, whereby the QM coder 13
The pixel of interest x _{i, j} is encoded according to the reference data.

As is clear from the above description, in this embodiment, the structural statistic of the image is obtained and the region determination is performed for each pixel of the image, and the pixel determination is performed for each pixel according to the region determination result for each pixel. Markov model coding is performed.
Therefore, the data compression rate is significantly improved as compared with the conventional JBIG encoding method that uses only the adjacent pixels that have already been encoded as reference data. Since the reference data for Markov model coding is switched according to the region determination data, it is optimal depending on whether it is a simple binary image such as a line figure or a pseudo halftone image by the tissue dither method or the error diffusion method. Encoding can be performed, and the overall data compression rate can be improved. Further, since it is possible to switch the reference data even in the middle of the line of the image, there is an advantage that the type of the input binary image does not matter.
Further, the calculation of the structural statistic of an image is a process that can be easily performed and requires a small amount of hardware as compared with the calculation of the correlation with the surrounding pixels and the calculation of the Markov model entropy for each pixel. Is. Therefore, the cost can be reduced as compared with the conventional method.

Next, the result of the simulation performed for this embodiment will be described. This simulation is a high-definition color standard image SCID (Standard Color
An image 512 obtained by reducing the portrait of (Image Data) to 1/16 and binarizing the luminance component by the error diffusion method.
X640 pixels and CCITT test chart No. 7
The image was obtained by cutting out 512 × 640 pixels from the above image and a mixed image of 1024 × 640 pixels was arranged side by side. When this mixed image for simulation was subjected to region determination, it was divided into two regions, a region that is a flat portion without a pixel period and a region that is a contour portion without a pixel period. Therefore, reference data in which the bit P _{i, j} representing the presence or absence of a pixel period is deleted from the above-mentioned 12-bit reference data is used. That is, this reference data is composed of 1 bit of area determination data, 4 bits of adaptive data, and 6 bits of fixed reference pixels, for a total of 11 bits. 2 using this reference data
^The Markov model consisting of ¹¹ (= 2048 ways) states was applied. That is, in order to verify the effectiveness of the present invention for mixed images, the number of states is limited to 2 ¹¹ ways.

Then, in the case of a conventional method using 11 pixels of proximity as reference data (model I), when 1 bit of area determination data and 10 pixels of proximity are used as reference data (model II), 1 bit of area determination data and proximity. For each of the case where 6 pixels and 4 bits of adaptive data are used as reference data (Model III),
The Markov model entropy H was calculated. Where the model
II and model III are within the scope of the invention.

[0044]

[Equation 4] In equation (3), S _i is the state of the reference data, x _j is the symbol of the binary image, P (S _i ) is the probability of appearance of the state S _i , and P (x _j |
S _i ) is the conditional probability that the symbol x _j appears in the state S _i . Markov model entropy H in each model
Were as follows:

Model I: H = 0.335 bits / pixel. Model II: H = 0.328 bits / pixel. Model III: H = 0.297 bits / pixel. Model I according to the conventional method and model according to the method of the present invention
By comparing the Markov model entropy H of II with each other, the effect of performing the region determination for each pixel and performing the state classification according to the result of the region determination was confirmed. Further, by comparing the Markov model entropy H of the model II and the model III, the effect of adaptively switching the reference data according to the region determination data was confirmed. According to the model III which adaptively generates the reference data based on the present invention, the Markov model entropy H can be reduced by about 11% as compared with the model I by the conventional method. This means that the compression rate of Markov model coding is improved by about 11%.

Next, a simulation was performed on an image having a periodic characteristic. The portrait of the above-described high-definition color standard image SCID is reduced to 1/16 and its luminance component is binarized by the tissue dither method.
An image of 0 pixels was created and used as an image for simulation. As a result of the area determination of the image for simulation, it was divided into two areas, that is, a flat area having a pixel cycle and an outline area having a pixel cycle. Therefore, as reference data, area determination data of 1 bit, adaptive data of 4 bits, and fixed reference pixel of 6 bits, which is a total of 11 bits, are used. As the area determination data, the data represented by A _{i, j} described above was used. Since the reference data is 11 bits, the number of states of the Markov model is 2 ¹¹ as in the above simulation. That is, in order to verify the effectiveness of the present invention for images having periodic characteristics, the number of states was limited to 2 ¹¹ ways.

Then, in the case of the conventional method using 10 adjacent pixels and 1 periodic pixel as reference data (model IV), 1 data bit for area determination, 6 adjacent pixels and 4 bits of adaptive data based on the method of the present invention. The Markov model entropy H was calculated by the equation (3) for each of the cases where is used as reference data (model V). However, as the adaptive data, 3-bit quantized data of one periodic pixel and the estimated density value L ^* _{i, j} was used. The results obtained are shown below.

Model IV: H = 0.176 bits / pixel. Model V: H = 0.160 bits / pixel. According to the model V according to the present invention, the Markov model entropy H can be reduced by about 9% as compared with the case of the model IV by the conventional method. This means that according to model V, the compression rate of Markov model coding is improved by about 9%. Further, according to the above-described embodiment, the number k of pixels used as the periodic pixels can be arbitrarily selected from 0 to. Therefore, when k is set to 2 (model V
I), that is, when the structure of the 4-bit adaptive data is 2-bit quantized data of two periodic pixels and the estimated density value L ^* _{i, j} , the Markov model entropy H was obtained.
As a result, H = 0.142 bits / pixel, and the model I
Compared with the conventional method represented by V, the model VI reduced Markov model entropy by about 19%. As described above, it was found that for an image having a periodic characteristic, the compression rate of Markov model coding is significantly improved by using several pixels having a strong correlation as adaptive data in the reference data.

[0049]

As described above, in the binary image data compression method of the present invention, the structural statistical amount of an image is obtained from the input binary image, and the area of the input binary image is determined according to this statistical amount. By creating the reference data in the Markov model coding in a form including the result of this area determination, adaptive Markov model coding according to the property of the image is performed, and Compared to the conventional case where only neighboring pixels are encoded as reference data,
There is an effect that a large amount of data compression can be realized with a small hardware scale. Further, the binary image data compression apparatus of the present invention is an image statistic calculating means for obtaining a structural statistic of an image from an input binary image, and an area for performing an area determination of the input binary image according to the statistic. By providing the determining means and the reference data generating means for creating the reference data in the Markov model encoding in the form including the result of the area determination, the adaptive Markov model encoding according to the property of the image is performed. As compared with the case of using the conventional data compression device, there is an effect that a large amount of data compression can be realized with a small amount of hardware.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a binary image data compression device according to an embodiment of the present invention.

FIG. 2 is a diagram showing coded adjacent 10 pixels used for variance calculation.

FIG. 3 is a block diagram showing a configuration of a distributed arithmetic unit.

FIG. 4 is a diagram showing an example of peripheral pixels used for calculation of an estimated density value.

FIG. 5 is a block diagram showing a configuration of an estimated density value calculation unit.

FIG. 6 is a block diagram showing a configuration of a period calculation unit.

FIG. 7 is a block diagram showing a configuration of a main part of an area determination unit.

FIG. 8 is a diagram showing an example of a reference data configuration in Markov data encoding.

FIG. 9 is a diagram showing an example of fixed reference pixels in reference data.

FIG. 10 is a block diagram showing a configuration of a main part of a reference data creation unit.

FIG. 11 is a block diagram showing a conventional binary image data compression device based on the JBIG method.

FIG. 12 is a diagram illustrating reference pixel positions in Markov model coding.

[Explanation of symbols]

 10 image statistic calculation unit 11 area determination unit 12 reference data creation unit 13 QM coder 14 variance calculation unit 15 estimated density value calculation unit 16 cycle calculation unit 21, 31, 33, 51 line memory 23, 32 adder 34 calculator 41 , 43, 44 Shift register 45, 46 Change point counter 52 Maximum value detection unit 53 Minimum value detection unit 54 Subtractor 57 Density region separation unit 61, 62 Quantizer 63 Selector 64 Adaptive data selection unit

Claims (14)

[Claims] 1. A data compression method of a binary image, the first step of obtaining an image statistic from an input binary image,
A second step of performing area determination of the input binary image according to the image statistics and generating area determination data, and a third step of generating reference data in Markov model coding
Of the binary image composed of the first data portion corresponding to the area determination data and the second data portion which is a data portion other than the first data portion. Data compression method. 2. The binary image data compression method according to claim 1, further comprising a fourth step of performing Markov model coding of the input binary image using the reference data. 3. The binary image data compression method according to claim 1, wherein the region determination is performed for each pixel of the input binary image. 4. The binary image data compression method according to claim 3, further comprising a fourth step of performing the Markov model coding of the pixel by distinguishing the Markov state of each pixel region using the reference data. 5. The second data portion is generated in the third step by selecting from a plurality of data according to the area determination data.
A method of compressing binary image data according to. 6. The binary image data compression method according to claim 5, wherein the second data portion is composed only of pixel values of coded pixels adjacent to the pixel of interest. 7. The pixel value of the encoded pixel having a strong correlation with the estimated density value of the pixel of interest calculated from the encoded pixel and the pixel value of the second data portion. And a pixel value of a pixel adjacent to the pixel of interest are combined, and the data compression method of the binary image according to claim 5. 8. The second data portion is a pixel value of a plurality of coded pixels having a strong correlation with the estimated density value of the pixel of interest calculated from the coded pixels and the pixel of interest. And a pixel value of a pixel adjacent to the pixel of interest are combined, and the data compression method of the binary image according to claim 5. 9. The binary image data compression method according to claim 3, wherein the area determination is for determining whether the pixel of interest is present in a contour portion or a flat portion of the image. 10. The binary image data compression method according to claim 3, wherein the region determination is for determining the presence or absence of a periodic structure in the input binary image. 11. A binary image data compression apparatus, comprising:
Image statistic calculating means for calculating an image statistic from an input binary image, area determining means for performing area determination of the input binary image according to the image statistic, and area determining means, and Markov model encoding And a first data portion in which the reference data corresponds to the area determination data and the first data portion.
And a second data portion that is a data portion other than the data portion of the binary image data compression apparatus. 12. The binary image data compression apparatus according to claim 11, further comprising an encoding unit that performs Markov model encoding of the input binary image using the reference data. 13. The binary image data compression apparatus according to claim 11, wherein the area determination means determines an area for each pixel of the input binary image. 14. The binary image data compression apparatus according to claim 13, further comprising coding means for distinguishing the Markov state of each pixel region using the reference data and performing Markov model coding of the pixel.