JPH08214159A - Image processing method and image processing unit - Google Patents

Abstract

PURPOSE: To obtain an excellent binary image by extracting a spatial frequency or a feature variable equivalent thereto for each picture element from a received image and changing a spread matrix in the error spread method or a threshold level matrix in the dither method based on the feature variable. CONSTITUTION: The image processing unit receiving a multi-gradation input image and providing an output of a binary image is provided with an image identification means 40 comprising a feature variable extract means 41 extracting a spatial frequency or a feature variable equivalent thereto based on a density of the multi-gradation input image and comprising a classification means 42 classifying the received image into some images based on the feature variable and providing an output of a classification signal corresponding to each classified image, and with a binarizing processing means 30 receiving the classification signal from the image identification means 40 and changing the binarizing processing method of the received image depending on the classification signal.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION The present invention relates to a printer, a scanner,
It is applied to copiers, facsimiles, etc.
The present invention relates to an image processing method and an image processing apparatus for converting into value image information.

[0002]

2. Description of the Related Art Conventionally, there are an error diffusion method and a dither method as methods for converting a multi-valued image into a binary image, which will be described below.

First, the error diffusion method will be described.

FIG. 17 is a block diagram for explaining the error diffusion method. In the figure, the multi-valued data d0 of the pixel of interest is read from the image memory 10 and the .gamma. Correction ROM 20 is referred to so that the input value d0 is .gamma.-corrected to match the characteristics of the output device such as a printer. The .gamma.-corrected multivalued data d0 'is corrected by the adder 31 of the binarization processing means 30 with the error Eij corresponding to the pixel of interest, and f = d0' + Eij is output.

Next, in the comparator 32, the binarization threshold Th is set.
When f ≧ Th, the binary signal B = “1”,
When f <Th, B = “0” is output. And 2
The subtractor 33 calculates the error E = f−B ′ due to the value conversion. Here, B'is a value obtained by multiplying B by 255 when the input multi-valued data d0 has 256 gradations (0 to 255). Therefore, for example, the binarization threshold value Th = 12
If f = 240, then B = “1”, but the error E due to binarization in this case is E = 240−1.
X becomes 255, and the error E due to binarization is E = -15
Is required. The binarization error E is error-distributed by the weighting error calculator 34 to the positions of pixels to be processed in the future using a predetermined error diffusion matrix, and stored in the error memory 35. Ie. In this case, when f = 240 is compared with the binarized threshold value Th = 128, the binarized signal is “1”, but in reality, the binarized signal “1” is 25.
255 in 6 gradations (0 to 255), f = 24
The value of 0 is 15 less than the value of 255 (-15). Therefore, for f = 240, the shortage of 15
Is used as an error, and this error is distributed to the error memory 35 of unprocessed pixels by the weighting error calculator 34 using the error diffusion matrix and reflected in the subsequent binarization.

An example of the error diffusion matrix is shown in FIG. FIG. 6 is a diagram for explaining the embodiment of the present invention, and the error diffusion will be described with reference to FIG.
In FIG. 6, the pixel indicated by * is the pixel of interest, and the error generated when the pixel of interest is binarized is distributed to the unprocessed pixels with the weight shown in FIG. For example, taking FIG. 6C as an example, “6” is given to the unprocessed pixel at this point,
The error is distributed with weights of “3”, “4”, and “3”. Then, when the binarization processing of the pixel of interest indicated by * is completed and the binarization processing of the next pixel is performed, the error value for the next pixel stored in the error memory 35 is read and the error value Is used to correct the input density value of the next pixel read from the image memory 10.

As described above, in the error diffusion method, after the binarization processing of a certain pixel is completed, the error generated in the binarization processing is distributed to the pixels to be processed thereafter, so that the overall error is obtained. The method is to reduce it.

In general, the quality of the processed image greatly changes depending on the error diffusion matrix used. Figure 6 (a)
When the error diffusion matrix with a wide distribution range is used, the part where the density distribution changes smoothly can be reproduced well, but the response property deteriorates in the part where the density distribution changes rapidly, and a very good result is obtained. Absent. In addition, since many sum of products operations are required, it takes a long processing time. On the other hand, FIG.
When the error diffusion matrix having a narrow distribution range shown in (c) is used, the reproducibility is good at the portion where the density changes abruptly. However, in the conventional error diffusion method, only one type of error diffusion matrix is used.

Next, the dither method will be described with reference to FIG. In FIG. 10A, each rectangle represents a pixel, and the pixel in the i-th column and the j-th row (the pixel shown by hatching) is the target pixel here and has the density value dij. . The dither method compares the input pixel density dij with the threshold value at the position of x and y corresponding to the i and j in the threshold value matrix Txy shown in FIG. ), The binarization is performed. That is, the output binary image Bij compares the input multi-valued image dij with the threshold value Txy of the threshold value matrix at the same position as the input multi-valued image dij, and when dij ≧ Txy, Bij =
Bij = "0" when "1" and dij <Txy.

In the threshold matrix used in the dither method, the calculation for (x, y) of the threshold Txy is performed at x = i% n and y = j% n. Here, n represents the matrix size (n = 4 in the case of FIG. 18 described above), and, for example, the operation a% b represents the remainder when the integer a is divided by the integer b. Therefore, x = i% n,
In y = j% n, when n = 4, the values of x and y are 0 to 3.

A concrete example of the threshold matrix used in the dither method is shown in FIG. FIG. 11 is a diagram for explaining the embodiment of the present invention. Generally, when the size of the threshold matrix is n × n, the number of gray scales that can be expressed is n ² +1. , The spatial resolution is 1 / n ² . Therefore,
When a matrix with a large n as shown in FIG. 11A is used, the number of gray scales that can be expressed increases, but the spatial resolution decreases. On the other hand, when a matrix with a small n is used as shown in FIG. 11C, the number of gray levels that can be expressed is small, but the spatial resolution is high. However, in the conventional dither method, only one threshold matrix is used.

[0012]

FIG. 19 is a diagram showing human visual characteristics. In FIG. 19, the horizontal axis represents the spatial frequency, and the vertical axis represents the number of gradations that can be normally recognized by humans. It can be said that the spatial frequency is high where black and white are inverted in a short cycle, and low when the number of black and white inversions is small and similar tones. As can be seen from FIG. 19, in the human visual characteristics, many gradations can be recognized at a low spatial frequency because the change in color tone is gentle, and at a high spatial frequency, there are extreme black and white. Because the change is noticeable,
The number of gray levels that can be recognized decreases.

However, in the conventional error diffusion method or dither method, the error diffusion matrix or threshold matrix used is fixed to one type, and it is impossible to change it so as to satisfy human visual characteristics. Met.

That is, the error diffusion method is as described above.
When the error diffusion matrix having a wide distribution range shown in FIG. 6A is used, the portion where the density distribution changes smoothly (the portion where the spatial frequency is low) can be reproduced well, but the portion where the density distribution changes rapidly (the space) In the high frequency part), the response characteristic is deteriorated and a very good result cannot be obtained.
When the error diffusion matrix having a narrow distribution range shown in (c) is used, the reproducibility is good at the portion where the concentration distribution changes rapidly (the portion where the spatial frequency is high), and it depends on the spatial frequency. An error diffusion matrix is required, but conventionally, the error diffusion matrix used is fixed to one type, and it has been impossible to change the color tone so as to satisfy human visual characteristics.

Also in the case of the dither method, as described above, when a threshold matrix with a large n as shown in FIG.
The spatial resolution becomes small, and when a threshold matrix with a small n is used as shown in FIG. 11C, the number of gray levels that can be expressed is small, but the spatial resolution becomes high. Although a threshold matrix is required, conventionally, the threshold matrix used is fixed to one type, and it has been impossible to change the color tone so as to satisfy human visual characteristics.

Therefore, in order to solve the above problems, the present invention extracts a spatial frequency or a feature amount corresponding thereto for each pixel from an input image, and according to the feature amount, an error diffusion method is used. Is to select the optimum error diffusion matrix, and to select the optimum threshold matrix in the dither method so as to obtain a good binary image that can satisfy the human visual characteristics. There is.

[0017]

The image processing method of the present invention comprises:
In an image processing method for inputting a multi-tone image and outputting a binary image, a feature amount based on density is extracted from the multi-tone input image, and the input image is classified into some according to the feature amount. Then, an image identification step of outputting a classification signal corresponding to each classification, and a binarization method that receives the classification signal from the image identification step and changes the binarization processing method of the input image according to the classification signal And a processing step.

As the feature amount based on the density in the image identifying step, the density gradient between the target pixel and its peripheral pixels is used, and the density gradient is the absolute value of the maximum density difference with the adjacent pixel or the density with the adjacent pixel. The average of the absolute values of the differences is used.

Further, the feature quantity based on the density in the image identifying step forms a block composed of a plurality of pixels including a target pixel, frequency-converts the pixel density included in the block, and calculates from a plurality of conversion coefficients. It is also possible to use the frequency feature value obtained by. Then, the frequency characteristic value is obtained by dividing the block composed of the plurality of transform coefficients into a plurality of sub-blocks, obtaining the sum of absolute values of the transform coefficients in the sub-blocks in each sub-block, It is derived by multiplying by a weighting factor.

Further, in the binarization process, in the case of the error diffusion method in which the binarization process is performed by using the error diffusion matrix, several error diffusion matrices are prepared so as to correspond to the classification of the image identification process. The error diffusion matrix to be used is prepared for each pixel according to the classification signal from the image identification step.

Further, in the case of the dither method in which the binarization process is performed by using the threshold value matrix in the binarization process step, several threshold value matrixes corresponding to the classification of the image identification process are used. Is prepared, and the threshold matrix to be used is selected for each pixel according to the classification signal in the image identification step.

The image processing apparatus of the present invention is an image processing method for inputting a multi-gradation image and outputting a binary image, wherein a feature amount based on density is extracted from the multi-gradation input image,
According to the classification signal, the input image is classified into some according to the feature amount, and the classification signal is output from the classification signal corresponding to each classification, and the classification signal from the image classification means is received. And a binarization processing unit that changes the binarization processing method of the input image.

As the feature amount based on the density in the image identifying means, the density gradient between the target pixel and its peripheral pixels is used, and the density gradient is the absolute value of the maximum density difference between adjacent pixels or the density between adjacent pixels. The average of the absolute values of the differences is used.

Further, the feature quantity based on the density in the image identifying means forms a block composed of a plurality of pixels including a target pixel, frequency-converts the pixel density included in the block, and calculates from a plurality of conversion coefficients. It is also possible to use the frequency feature value obtained by. Then, the frequency characteristic value is obtained by dividing the block composed of the plurality of transform coefficients into a plurality of sub-blocks, obtaining the sum of absolute values of the transform coefficients in the sub-blocks in each sub-block, It is derived by multiplying by a weighting factor.

Further, in the case of the error diffusion method in which the binarization processing means performs the binarization processing by using the error diffusion matrix, several error diffusion matrices are provided to correspond to the classification of the image identification means. The error diffusion matrix to be used is prepared for each pixel according to the classification signal from the image identification means.

Further, in the case of the dither method in which the binarization processing means uses the threshold matrix to perform the binarization processing, some threshold matrixes are provided so as to correspond to the classification of the image identification means. And the threshold matrix to be used is selected for each pixel according to the classification signal of the image identification means.

[0027]

As described above, according to the present invention, the spatial frequency or the characteristic amount corresponding to the spatial frequency is extracted from the multi-gradation input image based on the density, and the input image is classified into some according to the characteristic amount. Then, a classification signal corresponding to each classification is output, and the binarization processing method of the input image is changed according to the classification signal. Specifically, when the density gradient between the pixel of interest and the surrounding pixels is used as the feature amount based on the density, the absolute value of the maximum density difference between the adjacent pixels or the absolute value of the density difference between the adjacent pixels is used as the density gradient. The average value is used. Further, as a feature amount based on the density, a block composed of a plurality of pixels including a target pixel is formed, the pixel density included in the block is frequency-converted, and a frequency feature value calculated by a plurality of conversion coefficients is used. In this way, the spatial frequency or the feature amount corresponding to it is extracted based on the density for each input pixel, and the input image is classified into some in the frequency domain according to the feature amount, and the corresponding to each classification. Output the classification signal.

When the error diffusion method of performing the binarization processing using the error diffusion matrix is used as the binarization processing method, several error diffusion matrices are prepared to correspond to the classification of the image identification process. Then, the error diffusion matrix is selected for each pixel according to the classification signal from the image identification step. Further, when the dither method that performs the binarization processing using the threshold matrix is used as the binarization processing method,
Several threshold matrices are prepared to correspond to the classification of the image identification process, and the threshold matrix to be used is selected for each pixel according to the classification signal of the image identification process. In this manner, while distinguishing the image, the place where the density changes smoothly and the place where the density changes abruptly are discriminated, and the matrix suitable for it is selected to perform the binarization process. High-quality images that satisfy the characteristics can be reproduced.

[0029]

Embodiments of the present invention will be described below with reference to the drawings. Before describing the embodiments, the outline of the present invention will be first described.

FIG. 1 is a flow chart for explaining the outline of the overall processing of the image processing method of the present invention. In the figure, the image memory is sequentially scanned pixel by pixel, and a spatial frequency or a feature amount corresponding to the spatial frequency (this feature amount will be described later) is extracted from the density distribution of the target pixel and peripheral pixels (step s101). . Next, based on the extracted feature amount, it is classified whether the target pixel belongs to the low frequency region or the high frequency region (step s102). It should be noted that this classification is not divided into only two types of high and low frequency regions, but is equal to the number of matrices used in the binarization processing described later. Then, a matrix switching signal is transmitted according to the classification (step s1.
03) In the binarization process, the matrix used for binarization is selected for each pixel and binarized (step s
104).

FIG. 2 is a block diagram showing an overall schematic configuration of the image processing apparatus of the present invention. In FIG. 2, 10 is an image memory, 20 is a γ correction means, 30 is a binarization processing means, and 40 is. It is an image identification means composed of a feature quantity extraction means 41 and a classification means 42.

The multi-valued image signal read from the image memory 10 is γ-corrected by the γ-correction means 20 so as to match the characteristics of an output device such as a printer, and then sent to the binarization processing means 30. In the binarization processing means 30,
A matrix used in the binarization process is selected based on the matrix switching signal from the image identification means 40, the pixel density data is binarized using the selected matrix, and the binary image signal is output. To do.

The above is the outline of the present invention. Examples of the present invention will be described below.

(Embodiment 1) In Embodiment 1, an error diffusion method is used as a method for converting a binary image. Further, the spatial frequency or the characteristic amount corresponding to the characteristic amount has a density gradient between the pixel of interest and its peripheral pixels. First, as a density gradient, the density gradient between the pixel of interest to be processed and the adjacent pixel adjacent thereto is used. The case where the absolute value of the maximum density gradient (maximum density difference) of is used will be described.

FIG. 3 shows the image identifying means 40 in this case.
It shows the configuration of FIG. As described above, the image identifying means 40 is composed of the feature quantity extracting means 41 and the classifying means 42. Here, as the feature quantity, the maximum density gradient between the pixel of interest and the adjacent pixel adjacent thereto ( Since the absolute value of (maximum density difference) is used, the feature amount extraction means 4
1 has a density difference calculating means 411 for calculating a density difference between each adjacent pixel and a maximum density difference calculating means 412 for calculating a maximum density difference based on the calculated density difference.

FIG. 4 is a characteristic amount extraction process in the characteristic amount extraction means 41 shown in FIG. 3 (step s10 in FIG. 1).
1) and a detailed flowchart of the classification process (step s102 in FIG. 1) in the classification means. In this processing method, the spatial frequency is not directly obtained, but the characteristic amount corresponding to the spatial frequency is obtained by a simple calculation by utilizing the property that the density gradient becomes large in the high spatial frequency portion. . Hereinafter, description will be given along the flowchart of FIG.

First, in step s201, the maximum density difference d
Initialize max. Next, in step s202, the target pixel density d0 is input, and in step s203, the adjacent pixel density di is input. Here, the relationship between the pixel of interest and the adjacent pixel will be described with reference to FIG.

In FIG. 5, each rectangle represents one pixel, the pixel indicated by "0" is the pixel of interest currently being processed,
The pixels indicated by "1" to "8" are adjacent pixels. In the feature amount extraction processing, adjacent pixels are set to "1" to "4" (i = 1.
4 pixels, or "1" to "8" (i = 1 to 4)
It may be set to any of the 8 pixels of 8). When the number of adjacent pixels is four, the four pixels are adjacent pixels to be processed after the processing of the target pixel is completed. Therefore, in the raster scan method, in the case of FIG.
The pixels are "1" to "4". In performing the feature amount extraction process, it is slightly better to set the number of adjacent pixels to 8 pixels, but it is necessary to store the pixel density values for 3 lines in the image memory 10. On the other hand, when the number of adjacent pixels is four, the image memory 10 need only store pixel density values for two lines.
The capacity of 0 can be reduced, and the amount of calculation is halved. Practically enough accuracy is obtained even when the number of adjacent pixels is four.

Returning to the flowchart of FIG. 4, in step s204, the difference dif between the absolute values of the target pixel density d0 and each adjacent pixel density di is obtained. Next, in step s205, the density differences dif and dmax are compared, and dif
Is larger than dmax in step s206.
= Dif. That is, the difference in density value between each adjacent pixel is calculated and updated to the larger value. Then, in step s207, it is determined whether the comparison of the density differences between all the adjacent pixels is completed, that is, whether the number i of the adjacent pixels reaches i = imax, and i = i
If it is not max, it is determined that the comparison of the density differences with all the adjacent pixels is not completed, and in step s213, the number i of the adjacent pixel is incremented by 1 (i = i + 1) and the next adjacent pixel is determined. Enter the comparison. Here, the value of imax (maximum value of i) is im when the adjacent pixel to be referenced is 4 pixels.
ax = 4, im when the reference pixel is 8 pixels
ax = 8. The feature amount extraction processing is completed as described above, and the finally obtained dmax becomes the feature amount (maximum density difference) of the pixel of interest.

Next, dmax is compared with a preset threshold Th1, that is, it is determined whether dmax <Th1 (step s208). If dmax <Th1, the spatial frequency is in the low frequency region. And the switching signal is set to k = 1 (step s210), and dmax <T
When it is not h1, dmax is compared with a preset threshold value Th2 (Th2> Th1), that is, dm.
It is determined whether ax <Th2 (step s209).
In this determination, if dmax <Th2, the spatial frequency is considered to be in the middle frequency range, the switching signal k is set to 2 (step s311), and if dmax <Th2, the spatial frequency is considered to be in the high frequency range. Switching signal k =
3 (step s212).

Here, as an error diffusion matrix corresponding to each frequency region, there is an example as shown in FIG. FIG. 7A shows an example of the error diffusion matrix for the low frequency region, FIG. 8B shows an example of the medium frequency region, and FIG. 9C shows an example of the error diffusion matrix for the high frequency region. This error diffusion matrix shows an example in which the error is distributed to each pixel after the pixel of interest by the error diffusion method as described in the section of the prior art. Such an error diffusion matrix corresponding to each frequency region is selected by the switching signal k. That is, when k = 1, the error diffusion matrix Wxy (1) corresponding to the low frequency region of FIG. 9A is selected, and when k = 2, the error diffusion matrix Wxy (1) corresponding to the middle frequency region of FIG. 2) is selected, and when k = 3, the error diffusion matrix Wxy (3) corresponding to the high frequency region in FIG.

In this example, the feature amount (maximum density gradient) is classified into three types, but it is not necessary to limit to three types and the number of error diffusion matrices used in the binarization processing should be equal. Just classify. That is, in the above-described example, since the error diffusion matrix is set to three types, the feature amount (maximum density gradient) is classified into three stages correspondingly, but if the number of error diffusion matrices is increased, it corresponds to that. The feature amount (maximum concentration gradient) can be classified in multiple stages.

In the above process, the density gradient between adjacent pixels (here, the maximum density gradient) is used as the feature quantity, and the feature quantity corresponding to the spatial frequency is obtained based on the magnitude of the maximum density gradient. This is a process of determining what frequency region the input image belongs to and outputting a switching signal for selecting an error diffusion matrix corresponding to the frequency region. The conversion processing will be described later.

Next, as the feature amount, the average value of the absolute values of the density gradients with the adjacent pixels is used, and the feature amount corresponding to the spatial frequency is obtained based on the magnitude of the average value. An example will be described in which it is determined whether the signal belongs to a certain frequency region and a process of outputting a switching signal for selecting an error diffusion matrix corresponding thereto is performed.

In the present processing method, as in the above, the spatial frequency is not directly obtained, but in the high spatial frequency portion,
By utilizing the property that the density gradient becomes large, the feature quantity corresponding to the spatial frequency is obtained by a simple calculation. Less than,
This will be described with reference to the flowchart of FIG. In addition,
The configuration of the image identifying means 40 for realizing the present processing method is that the maximum density difference calculating means 412 in the image identifying means 40 described with reference to FIG. Omit it.

In the flowchart of FIG. 7, first, in step s301, the average density difference dave is initialized (d
After ave = 0), the target pixel density d0 to be processed is input (step S302). Next, the density di of the adjacent pixel adjacent to the pixel of interest is input (step S3).
03). The relationship between the target pixel and the adjacent pixel is as described above with reference to FIG.

Then, the difference dif between the absolute values of the density gradients of the target pixel density d0 and the adjacent pixel density di is obtained (step S304). Then, the difference dif of the absolute value of the density gradient obtained at that time is added to dave initialized (= 0) in step S301 (step S305), and the comparison of the density difference between all adjacent pixels is completed. It is determined whether or not the number i of the adjacent pixel has reached i = imax. If i = imax is not satisfied, it is determined that the comparison of the density differences between all the adjacent pixels has not been completed. , The number i of the adjacent pixel is incremented by 1 (step s31
3), the steps S303 to s between the next adjacent pixel
The process of 306 is performed. Here, as described above, the value of imax is i when the number of adjacent pixels to be referred to is four.
max = 4, imax = 8 when the adjacent pixels to be referred to are eight pixels.

When the comparison of the density differences between all the adjacent pixels is completed and the respective density differences are added, the added density difference dave is divided by imax to obtain the average density gradient dave. Obtained (step S307). With this, the feature amount extraction processing ends, and in this case, the average density gradient dave becomes the feature amount of the pixel of interest.

Next, the average density gradient dave thus obtained and the preset threshold value Th1
1 is compared, that is, it is determined whether or not dave <Th11 (step s308), and if dave <Th11,
The density gradient is small on average, the spatial frequency is considered to be in the low frequency region, and the switching signal is set to k = 1 (step s
310). If dave <Th11 is not satisfied, d
ave is compared with a preset threshold Th21 (where Th21> Th11), that is, dave <T
It is determined whether or not h21 (step s309), and dave
If <Th21, the spatial frequency is considered to be in the middle frequency region, and the switching signal is set to k = 2 (step s31
1) If not dave <Th21, the concentration gradient is large on average, the spatial frequency is regarded as a high frequency region, and the switching signal is set to k = 3 (step s312).

Then, as described above, when k = 1, FIG.
Error diffusion matrix Wxy corresponding to the low frequency region of
(1) is selected, when k = 2, the error diffusion matrix Wxy (2) corresponding to the middle frequency region of FIG. 11B is selected, and when k = 3, the error diffusion matrix Wxy (2) corresponds to the high frequency region of FIG. The error diffusion matrix Wxy (3) is selected.

In this example, the feature amount (average density gradient) is classified into three types, but it is not necessary to limit the number to three types, and the number of error diffusion matrices used in the binarization process should be equal. Just classify. That is, in the above example, since there are three types of error diffusion matrices,
Correspondingly, the feature amount (average density gradient) is classified into three levels, but if the number of error diffusion matrices is increased, the feature amount (average density gradient) may be classified into multiple levels correspondingly. it can.

In the above processing, the average density gradient between the adjacent pixels is used as the feature quantity, and the feature quantity corresponding to the spatial frequency is obtained based on the magnitude of the average density gradient to determine what the input image is. This is a process of determining whether the error diffusion matrix belongs to the frequency domain and outputting a switching signal for selecting an error diffusion matrix corresponding to it, and the selection switching process and the binarization process of the error diffusion matrix by this switching signal will be described below. Explained.

FIG. 8 is a block diagram showing a hardware configuration for implementing the error diffusion matrix switching process and the binarization process using the error diffusion method. The configuration shown in FIG. 8 is the same as that of FIG. 2 as a whole, and includes an image memory 10, a γ correction ROM 20 as a γ correction means, a binarization processing means 30,
The image identification means 40 is used. Although not shown here, the image identifying means 40 is composed of a feature quantity extracting means 41 and a classifying means 42, and the feature quantity extracting process and the classifying process are as described with reference to the flowcharts of FIGS. 4 and 7. Here, the binarization processing means 30 will be described.

The binarization processing means 30 has an adder 31, a comparator 32, a subtractor 33, a weighting error calculator 34, and an error memory 35, as in FIG. 17 used in the description of the conventional example. Further, in the first embodiment of the present invention, in addition to these, the error diffusion matrix selecting means 36 is provided.

In such a configuration, the image memory 10
Read the multivalued data d0 of the pixel of interest from
With reference to the OM 20, the input value d0 is γ-corrected to d0 ′ so as to match the characteristics of the output device such as a printer. The γ-corrected multivalued data d0 ′ is corrected by the adder 31 of the binarization processing unit 30 with the error Eij corresponding to the pixel of interest, and f
= D0 '+ Eij is output. This f is the comparator 32
Then, it is compared with the binarization threshold Th, and the binary signal B = “1” is output from the comparator 32 when f ≧ Th, and B = “0” is output when f <Th. And the error E due to binarization
= F−B ′ is calculated by the subtractor 33. Here, B'is the input multi-valued data d0 having 256 gradations (0
In the case of ~ 255), the value is obtained by multiplying B by 255. Therefore, for example, assuming that the binarization threshold value Th = 128, B = 1 at f = 240, but the error E due to binarization in this case becomes E = 240-1 × 255. The error E due to binarization is calculated as E = -15.

On the other hand, the error diffusion matrix selecting means 36
Outputs the designated error diffusion matrix Wxy (k) to the weighting error calculator 34 based on the matrix switching signal (k = 1 to 3) sent from the image identifying means 40. The weighting error calculator 34 sets the target pixel E based on the designated error diffusion matrix Wxy (k).
The error Ei'j 'to be distributed to the unprocessed pixels is calculated [Ei'j' =
Wxy (k) · E], the error Ei′j ′ is distributed to the position of the pixel to be processed in the future and stored in the pixel error memory 35.

FIG. 9 is a flow chart for explaining the above processing. In FIG. 9, first, the target pixel density d0 which is a binarization processing target is read from the image memory 10 (step S401), and γ correction (d0 → d0 ′) is performed according to the characteristics of an output device such as a printer. Perform (step S402). Next, the error Eij corresponding to the pixel of interest is read from the error memory 35, and the error correction (fij =
d0 '+ Eij) is performed (step S403). The data fij after the error correction is compared with the binarization threshold Th (step S404). If fij is larger than Th, the binarization output B = “1”, and if it is smaller, B = “0”. Output. Then, the error E caused by binarization is calculated (E = f
ij-B ') (step S405). Here, when the input pixel density d0 is 256 gradations (value range is 0 to 255), B ′ = B × 255 as described above. Then, according to the matrix switching signal (k = 1 to 3) sent from the image identification processing, the error diffusion matrix Wxy
Select (k). An example of the error diffusion matrix used here is shown in FIG. FIG. 6A shows a low frequency region (k =
1) The processing matrix has a wide error diffusion range, so that smooth image quality can be reproduced. On the other hand, FIG.
Since this is a matrix for high-frequency region (k = 3) processing and has a narrow error diffusion range, it is possible to excellently reproduce an edge region in which the density sharply changes. Further, FIG. 6B is a matrix for processing the medium frequency region (k = 2).

Then, in step s407, the error Ei'j 'to be distributed is calculated based on the selected error diffusion matrix Wxy (k) (Ei'j' = Wxy (k) .E), The error Ei'j 'is distributed to the position of the pixel to be processed in the future and stored in the error memory 35 (step s408).

As described above, according to the method of the present invention, while distinguishing an image, a place where the density changes smoothly and a place where the density changes abruptly are discriminated, and an error diffusion matrix suitable for it is selected to perform the binarization process. Since it is performed, it is possible to reproduce a high-quality image that satisfies the human visual characteristics.

This is the end of the description of the example using the error diffusion method.

(Embodiment 2) Next, as Embodiment 2 of the present invention, a case where a dither method is used as a binarization processing method will be described. In addition, the part other than the binarization processing method, that is,
The process until the feature amount corresponding to the spatial frequency is obtained and the matrix switching signal corresponding to the feature amount is output is the same as that in the first embodiment, and therefore the description of the process procedure up to this point is omitted here. To do.

Here, the threshold value matrix used in the dither method is the matrix switching signal (k = 1 to 1).
The process of selecting 3) will be described.

FIG. 10 is a block diagram showing a hardware configuration for implementing threshold value matrix switching processing using the dither method and binarization processing using this threshold value matrix. The configuration shown in FIG. 10 is the same as that of FIG. 2 as a whole, and includes an image memory 10, a γ correction unit 20, a binarization processing unit 30, and an image identification unit 40. The feature amount extraction processing and the classification processing by the image identification unit 40 are as described with reference to the flowcharts of FIGS. 4 and 7, and the binarization processing unit 30 will be described here.

Binarization processing means 30 by this dither method
Is a threshold matrix ROM 301, column counter 30
2, a row counter 303, and a comparator 304.

The threshold matrix ROM 301 is
For example, the threshold matrix Txy (1) for the low frequency region as shown in FIG. 11 (a), the threshold matrix Txy (2) for the medium frequency region as shown in FIG. 11 (b), A threshold matrix Txy (3) for a high frequency area as shown in (c) is stored. Also, the column counter 302
The row counter 303 counts up each time the pixel of interest is scanned from the image memory 10 and outputs the count value (i, j). However, if the values n of x and y in the largest size matrix are n = 8, 0 to 7 are repeatedly output from the column counter 302 and the row counter 303.

The threshold matrix ROM 301 has a matrix switching signal (k = 1 to 1) from the image identifying means 40.
3), the outputs (i, j) from the column counter 302 and the row counter 303 are addressed and input. For example, if the matrix switching signal is k = 1 and the column counter 30
It is assumed that 2 and the output (i, j) from the row counter 303 are both 3. In this case, the threshold matrix Txy (1) of FIG. 11A is selected, and as can be seen from the figure,
The threshold value “18” is output to the comparator 304 from the position of x, y corresponding to i, j = “3”. The comparator 304 compares this threshold value with the γ-corrected input pixel density, and if the input pixel density is larger than the threshold value, “1”,
If it is smaller, a binary signal of "0" is output.

By the way, the position of the input pixel, that is, the output (i, j) from the column counter 302 and the row counter 303.
From the corresponding threshold matrix Txy (K) (k = 1 to
When deriving the position (x, y) of 3), if the size n of the threshold matrix Txy (K) used changes, the calculation becomes slightly complicated. In order to prevent this, for example, as in the second embodiment, the threshold matrix T for the low frequency region is set.
The matrix size of xy (1) is n = 8, the threshold matrix Txy (2) for the medium frequency region is n = 4, and the threshold matrix Txy (3) for the high frequency region is n.
= 2, the threshold matrix Txy for the low frequency region
(1) Based on the size (n = 8), for the medium frequency region, as shown in FIG. 12 (a), n = shown in FIG. 11 (b).
The matrix of n = 8 is formed by arranging the matrix of 4 in rows of 2 vertically and horizontally. For the high frequency region, the matrix of n = 2 shown in FIG.
By forming a matrix of n = 8 arranged one by one, (x, y) can always be calculated from the value of (i, j) with n = 8.

FIG. 13 is a flow chart for explaining the binarization processing in the dither method. In FIG. 13, first,
The target pixel density d0 (i, j) is read (step s50
1), γ correction is performed and d0 ′ is output (step s50).
2). Next, the threshold matrix Txy (k) (k = 1 to 3) is selected according to the matrix switching signal (k = 1 to 3) sent from the image identifying means 40 (step s503). Then, the threshold values at the positions of x and y corresponding to the positions of i and j of the selected threshold value matrix are extracted and compared with the γ-corrected density value d0 ′ of the input pixel to obtain a binary value. A signal is output (step s504).

As described above, also in the second embodiment, while the image is being discriminated, the place where the density changes smoothly and the place where the density changes abruptly are discriminated, and the threshold matrix suitable for it is selected and binarized. I am trying to process it.
That is, the threshold matrix for the low frequency region shown in FIG. 11A can express 65 gradations, but since the spatial resolution is low, it is suitable for processing the portion where the density distribution changes smoothly. Although the threshold matrix for high frequency region processing shown in (c) can express only 5 gradations, it has a high spatial resolution and is suitable for processing a portion where the density distribution changes abruptly. Since the threshold matrix is selected according to the feature amount of the input pixel and the binarization process is performed by the selected threshold matrix, it is possible to reproduce a high quality image that satisfies the human visual characteristics.

(Third Embodiment) In the first and second embodiments described above, the density gradient between adjacent pixels is calculated as the feature quantity corresponding to the spatial frequency used in the image identification processing, and the density is calculated. The matrix for binarization is selected based on the gradient, but in the third embodiment, the case where the frequency conversion is used for the feature amount extraction used in the image identification processing will be described.

The third embodiment differs from the first and second embodiments in the structure and processing of the image identifying means, and is otherwise the same as the first or second embodiment. Note that there are various methods such as fast Fourier transform (FFT) and discrete cosine transform (DCT) as the frequency conversion method that can be used in the third embodiment, but in this embodiment, the Hadamard transform that is easy to calculate is used. The example used will be described.

FIG. 14 shows the configuration of the image identifying means 40 in the third embodiment, which is composed of a frequency converting means 511, a sub-block sum total calculating means 512, and a maximum sub-block detecting means 513. The block sum calculation means 512 corresponds to the feature amount extraction means 41 described in the first embodiment, and the maximum sub-block detection means 513 corresponds to the classification means 42 described in the first embodiment.

The frequency conversion means 511 is used for the image memory 10
Then, the pixel density data of the n × n block including the pixel of interest is read, frequency conversion is performed by two-dimensional Hadamard conversion, and a conversion matrix is obtained. Sub-block sum calculation means 5
12 calculates the sum S of the absolute values of the transform coefficients for each subblock from the transform matrix. Maximum sub-block detection means 513
Detects the sub-block for which the sum S of absolute values is maximum,
The corresponding matrix switching signal k is transmitted. Note that these will be described in detail below with reference to the flowchart in FIG.

In FIG. 15, n × n including the target pixel
(In Hadamard transform, n is a power of 2) blocks are formed (step s601), and the densities of these pixels are read from the image memory 10. Then, the pixel density in the block is two-dimensionally Hadamard transformed to obtain a transformation matrix (step s602). This conversion matrix is obtained as an 8 × 8 integer matrix when it is used for an 8 × 8 pixel block. Next, this conversion matrix is divided into a plurality of sub-blocks. FIG. 16 shows an example of division into these sub-blocks.
In this example, three sub blocks sb1, sb2, sb
Divided into three. In the transform coefficient of the Hadamard transform, the low-frequency component appears at the position a at the upper left in FIG. 16, and the high-frequency component appears at the position b at the lower right. You can understand what kind of characteristics they have. That is, whether the density distribution of the block including the pixel of interest belongs to a high frequency region or a low frequency region in terms of spatial frequency,
Further, it can be grasped whether it belongs to the medium frequency region. Note that FIG. 16 shows an example in which the sub-blocks are divided into straight lines for convenience of description, but the sub-blocks are not necessarily divided into straight lines in practice.

Next, each of the sub-blocks sb1 and s
sum s1, s of absolute values of the transformation matrix in b2, sb3
2, s3 are obtained (step s604). Then, the sum s1, s of the absolute values of the conversion matrix for each of these sub blocks
2, s3 is multiplied by a weighting coefficient Ak (k = 1, 2, 3) to obtain feature quantities A1 · s1, A2 in each sub-block.
-S2, A3, s3 are obtained (step s605). Multiplying the weighting coefficient A here generally means that in the Hadamard transform matrix, there is a phenomenon that the value of the low frequency component is large and the value of the high frequency component is small, and the number of transform coefficients in each sub-block is This is to correct that the sub-blocks are not equal. Next, the feature quantity A1 · s1, A
The sub-block having the maximum 2 · s2 and A3 · s3 is detected (step s606). Then, the matrix switching signal k (k = 1 to 3) corresponding to the sub-block for which the feature amount is determined to be the maximum is output (step s607). Subsequent matrix selection processing can be applied to both the first and second embodiments.

At the steps s606 and 607, for example, the characteristic amounts A1.s1, A2.s2 of the sub-blocks which form the block including the pixel of interest.
If A1 · s1 is the largest of A3 · s3,
This means that the low frequency component is large, and it can be determined that the block belongs to the low frequency region. In this case, k = 1 is output as the matrix switching signal. Also, A2 ・ s
When 2 is the largest, it means that the medium frequency component is large, and it can be determined that the block belongs to the medium frequency region. In this case, k = 2 is output as the matrix switching signal. Further, when A3 · s3 is the largest, it means that the high frequency component is large, and it can be determined that the block belongs to the high frequency region. In this case, k = 3 is output as the matrix switching signal. The weighting factor and the method of dividing the sub-blocks are determined experimentally by obtaining a transformation matrix for various samples.

As described above, in the third embodiment, since the characteristic amount of the density distribution of the input pixel is obtained by frequency conversion,
A highly accurate feature amount can be obtained. Therefore, when selecting a matrix for binarization according to the feature amount of the input pixel, a more accurate matrix can be selected, and the binarization processing is performed by the matrix that is selected accurately. It is possible to reproduce high-quality images that satisfy the human visual characteristics.

[0078]

As described above, in the image processing method according to the present invention, according to the first aspect, the feature quantity corresponding to the spatial frequency is extracted from the multi-tone input image based on the density, and the feature thereof is extracted. The input image is classified into several according to the amount,
Since the classification signal corresponding to each classification is output and the binarization processing method of the input image is changed according to this classification signal, a high-quality binarized image satisfying human visual characteristics can be obtained. Therefore, it is possible to greatly contribute to the improvement of image quality in printers and copiers.

Further, according to the second aspect, by using the density gradient with surrounding pixels as the feature quantity corresponding to the spatial frequency based on the density, the processing for obtaining the feature quantity is simple, and the processing time is short. Processing becomes possible and the processing speed can be increased.

According to the third aspect, by using the absolute value of the maximum density difference between adjacent pixels as the density gradient,
Since the edge portion of the image can be detected sensitively, it is suitable for obtaining sharp image quality.

According to the fourth aspect, by using the average value of the absolute values of the density differences with the adjacent pixels as the density gradient, the density differences in all directions are averaged, so that a soft image quality is obtained. It is suitable for.

According to the fifth aspect, as the feature amount based on the density, a block composed of a plurality of pixels including the target pixel is formed, the pixel density included in the block is frequency-converted, and a plurality of conversion coefficients are obtained. The spatial frequency can be accurately detected by using the frequency feature value obtained by the calculation from, and thus the frequency range to which the input pixel belongs can be accurately classified, resulting in higher quality. Binarization processing is possible.

According to a sixth aspect of the present invention, the block consisting of the plurality of transform coefficients is divided into a plurality of sub-blocks, and in each sub-block, the sum of absolute values of the transform coefficients in the sub-blocks is calculated. Since it is derived by multiplying each absolute value sum by a weighting coefficient for correcting different values between sub-blocks, it is possible to easily and accurately classify which frequency region an input pixel belongs to. . Further, according to claim 7, when the error diffusion method for performing the binarization processing using the error diffusion matrix is used as the binarization processing method, some error diffusions are performed to correspond to the classification of the image identification process. A matrix is prepared, and an error diffusion matrix is selected for each pixel according to the classification signal from the image identification process. That is, while distinguishing the image, the place where the density changes smoothly and the place where the density changes abruptly are discriminated, and the binarization process is performed by selecting the error diffusion matrix suitable for it, so that the human vision High-quality images that satisfy the characteristics can be reproduced. Further, by using the error diffusion method as the method of performing the binarization processing, more accurate binarization processing can be performed, and high quality image quality can be obtained.

According to the eighth aspect, when the dither method for performing the binarization process using the threshold value matrix is used as the binarization process method, several methods are used to correspond to the classification of the image identification process. The threshold value matrix is prepared, and the threshold value matrix to be used is selected for each pixel according to the classification signal in the image identification process. That is, while distinguishing the image, the place where the density changes smoothly and the place where the density changes abruptly are discriminated, and the threshold value matrix suitable for it is selected to perform the binarization process.
It is possible to reproduce high-quality images that satisfy the human visual characteristics. Further, by using the dither method as the method of performing the binarization processing, the processing can be simplified and the processing time can be shortened.

Further, in the image processing apparatus of the present invention, according to the ninth aspect, the feature quantity corresponding to the spatial frequency is extracted from the multi-tone input image based on the density, and the input image is extracted according to the feature quantity. Are classified into several groups, a classification signal corresponding to each classification is output, and the binarization processing method of the input image is changed according to this classification signal, so that the human visual characteristics can be satisfied. It is possible to obtain a high quality binarized image, which can greatly contribute to the improvement of image quality in a printer, a copier, or the like.

According to the tenth aspect, by using the density gradient with surrounding pixels as the feature quantity corresponding to the spatial frequency based on the density, the processing for obtaining the feature quantity is simple and requires a short processing time. Processing becomes possible and the processing speed can be increased.

According to the eleventh aspect, by using the absolute value of the maximum density difference with the adjacent pixel as the density gradient, the edge portion of the image can be detected sensitively, so that a sharp image quality can be obtained. It is suitable for.

According to the twelfth aspect, since the average of the absolute values of the density differences with the adjacent pixels is used as the density gradient, the density differences in all directions are averaged, so that a soft image quality is obtained. It is suitable for.

According to the thirteenth aspect, as the feature quantity based on the density, a block composed of a plurality of pixels including the target pixel is formed, the pixel density included in the block is frequency-converted, and a plurality of conversion coefficients are obtained. The spatial frequency can be accurately detected by using the frequency feature value obtained by the calculation from, and thus the frequency range to which the input pixel belongs can be accurately classified, resulting in higher quality. Binarization processing is possible.

According to a fourteenth aspect, the block consisting of the plurality of transform coefficients is divided into a plurality of sub-blocks, and in each sub-block, the sum of absolute values of transform coefficients in the sub-blocks is calculated, Since it is derived by multiplying each absolute value sum by a weighting coefficient for correcting different values between sub-blocks, it is possible to easily and accurately classify which frequency region an input pixel belongs to. . According to claim 15,
When the error diffusion method of performing the binarization processing using the error diffusion matrix is used as the binarization processing method, some error diffusion matrices are prepared to correspond to the classification of the image identification process, and the image identification means is used. The error diffusion matrix is selected for each pixel according to the classification signal of. That is, while distinguishing the image, the place where the density changes smoothly and the place where the density changes abruptly are discriminated, and the binarization process is performed by selecting the error diffusion matrix suitable for it, so that the human vision High-quality images that satisfy the characteristics can be reproduced. Further, by using the error diffusion method as the method of performing the binarization processing, more accurate binarization processing can be performed, and high quality image quality can be obtained.

According to the sixteenth aspect, when the dither method for performing the binarization processing using the threshold value matrix is used as the binarization processing method, there are several methods corresponding to the classification of the image identification means. The threshold value matrix is prepared, and the threshold value matrix to be used is selected for each pixel according to the classification signal of the image identifying means. That is, while distinguishing the image, the place where the density changes smoothly and the place where the density changes abruptly are discriminated, and the threshold value matrix suitable for it is selected to perform the binarization process. High-quality images that satisfy visual characteristics can be reproduced. Further, by using the dither method as a method of performing the binarization processing, the processing can be simplified,
The processing time can be shortened.

[Brief description of drawings]

FIG. 1 is a flowchart illustrating the outline of the overall processing of the present invention.

FIG. 2 is a block diagram illustrating the outline of the overall configuration of the present invention.

FIG. 3 is a diagram showing a configuration example of an image identifying means in the first embodiment of the present invention.

FIG. 4 is a flowchart illustrating processing of an image identification unit when a maximum density difference is used as a feature amount in the first embodiment.

FIG. 5 is a diagram illustrating a relationship between a pixel of interest and an adjacent pixel.

FIG. 6 is a diagram showing an example of an error diffusion matrix used in the first embodiment.

FIG. 7 is a flowchart illustrating processing of an image identification unit when an average value of density differences is used as a feature amount in the first embodiment.

FIG. 8 is a diagram showing an overall configuration according to the first embodiment, particularly a diagram for explaining the configuration of the binarization processing means.

FIG. 9 is a flowchart illustrating processing of the binarization processing unit according to the first exemplary embodiment.

FIG. 10 is a diagram showing an overall configuration according to a second embodiment of the present invention, particularly a diagram for explaining the configuration of the binarization processing means.

FIG. 11 is a diagram showing an example of a threshold matrix used in the second embodiment.

FIG. 12 is a diagram illustrating a method for making the matrix sizes common in the threshold matrix used in the second embodiment.

FIG. 13 is a flowchart illustrating a process of a binarization processing unit according to the second exemplary embodiment.

FIG. 14 is a diagram showing a configuration example of an image identifying means in a third embodiment of the present invention.

FIG. 15 is a flowchart illustrating a process of image identifying means according to the third embodiment.

FIG. 16 is a diagram showing an example in which the frequency conversion matrix in the third embodiment is divided into a plurality of blocks.

FIG. 17 is a block diagram showing a schematic configuration of a conventional image processing apparatus using an error diffusion method as a binarization processing method.

FIG. 18 is a diagram for explaining a dither method as a binarization processing method.

FIG. 19 is a diagram showing the relationship between the spatial frequency and the number of recognizable gradations for explaining human visual characteristics.

[Explanation of symbols]

 10 ... Image memory 20 ... γ correction means 30 ... Binary processing means 34 ... Weighting error calculator 35 ... Error memory 36 ... Error diffusion matrix selection means 40 ... Image Identification means 41 ... Feature amount extraction means 42 ... Classification means 301 ... Threshold matrix ROM 302 ... Column counter 302 ... Row counter 411 ... Density difference calculation means 412 ... Maximum Density difference calculation means 511 ... Frequency conversion means 512 ... Sub block total sum calculation means 513 ... Maximum sub block detection means

Claims (16)

[Claims] 1. An image processing method for inputting a multi-tone image and outputting a binary image, wherein a feature amount based on density is extracted from the multi-tone input image, and the input image is extracted according to the feature amount. Into several categories,
An image identification step of outputting a classification signal corresponding to each classification, and a binarization processing step of receiving a classification signal from the image identification step and changing a binarization processing method of an input image according to the classification signal. An image processing method comprising: 2. The image processing method according to claim 1, wherein the feature amount based on the density in the image identifying step is obtained based on the density gradient between the pixel of interest and its peripheral pixels. 3. The image processing method according to claim 2, wherein an absolute value of a maximum density difference between adjacent pixels is used as the density gradient. 4. The image processing method according to claim 2, wherein the density gradient uses an average value of absolute values of density differences between adjacent pixels. 5. The feature quantity based on the density in the image identifying step forms a block composed of a plurality of pixels including a target pixel, frequency-converts the pixel density included in the block, and calculates from a plurality of conversion coefficients. The image processing method according to claim 1, wherein the obtained frequency feature value is used. 6. The frequency characteristic value is obtained by dividing a block composed of the plurality of transform coefficients into a plurality of sub-blocks,
In each sub-block, the sum of absolute values of the transform coefficients in the sub-block is obtained, and in order to correct different values between the sub-blocks, the sum of absolute values is multiplied by a weighting factor to derive the sum. The image processing method according to claim 5. 7. The error diffusion method in which the binarization process is performed by using an error diffusion matrix,
7. A number of error diffusion matrices are prepared to correspond to the classification of the image identification process, and the error diffusion matrix to be used is selected for each pixel according to the classification signal from the image identification process. 1. The image processing method described in 1. 8. The binarization processing step is a dither method in which the binarization processing is performed using a threshold value matrix, and several threshold value matrixes correspond to the classification of the image identification step. The image processing method according to claim 1, wherein the threshold value matrix to be used is selected for each pixel according to the classification signal of the image identification step. 9. An image processing apparatus for inputting a multi-gradation image and outputting a binary image, comprising: a feature amount extraction means for extracting a feature amount based on density from the multi-tone input image; In response to the classification signal output from the image classification means, which has classification means for classifying the input image into several and outputting classification signals corresponding to the classifications, and receiving the classification signal output from the image classification means. And a binarization processing unit for changing the binarization processing method of the input image. 10. The image processing apparatus according to claim 9, wherein the feature amount based on the density in the image identifying means is obtained based on a density gradient between a target pixel and its peripheral pixels. 11. The image processing apparatus according to claim 10, wherein the density gradient uses an absolute value of a maximum density difference between adjacent pixels. 12. The density gradient uses an average value of absolute values of density differences between adjacent pixels.
The image processing device according to item 1. 13. The feature quantity based on the density in the image identifying means forms a block composed of a plurality of pixels including a target pixel, frequency-converts the pixel density included in the block, and calculates from a plurality of conversion coefficients. The image processing apparatus according to claim 9, wherein the frequency characteristic value obtained by the above is used. 14. The frequency characteristic value is obtained by dividing a block composed of the plurality of transform coefficients into a plurality of sub-blocks, obtaining the sum of absolute values of the transform coefficients in the sub-blocks in each sub-block, 14. The image processing apparatus according to claim 13, wherein the sum of absolute values is derived by multiplying by a weighting coefficient in order to correct different values between sub-blocks. 15. The binarization processing means is an error diffusion method for performing binarization processing using an error diffusion matrix, wherein several error diffusion matrices are provided to correspond to the classification of the image identification means. The image processing apparatus according to claim 9, wherein the error diffusion matrix to be used is prepared and selected for each pixel according to the classification signal from the image identification means. 16. The dithering method is a dither method for performing binarization processing using a threshold value matrix,
Some threshold matrices are prepared to correspond to the classification of the image identifying means, and the threshold matrix to be used is selected for each pixel according to the classification signal of the image identifying means. Item 9. The image processing device according to item 9.