JPH07264397A - Image picture processor - Google Patents

Abstract

PURPOSE:To separate characters and photographs without erroneously recognizing a part where density change is rapid in a photograph part or the like and to discriminate whether a character part is black characters on a white base or white characters on a black base. CONSTITUTION:A block black and white representative picture element value detection part 1 obtains the representative value of white picture elements and the representative value of black picture elements for the respective blocks of multilevel images divided into the blocks. An inter-block deviation integration part 2 calculates the sum of the difference among the blocks of the representative values of the white picture elements and the black picture elements obtained by the block black and white representative picture element value detection part 1. A character image identification part 3 obtains a feature amount for expressing whether the block is the characters or the images from the value of the sum of the difference of the representative values of the white picture elements and the black picture elements among the blocks obtained by the inter-block deviation integration part 2. In addition, a character background identification part for discriminating the black characters in white background or the white charactors in black background for the blocks discriminated as the characters by the character image identification part 3 is provided.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing apparatus,
More specifically, the present invention relates to an image processing apparatus that adaptively binarizes by identifying whether an area on a multi-valued image is a character or an image, and if it is a character, it is black on a white background or white on a black background.

[0002]

2. Description of the Related Art As a conventional image processing device, a device has been proposed in which an image is separated into a character portion, a photograph portion, a halftone dot portion, and the like. For example, in Japanese Patent Laid-Open No. 62-147860, the input pixels are divided into blocks, and if the difference between the maximum value and the minimum value in the block is larger than a certain threshold value T1, the character / halftone dot portion is displayed. And If the number of positive / negative changes in the difference between two consecutive pixels is larger than the threshold value T2, the halftone dot portion is selected, and if it is smaller, the character portion is selected. In Japanese Patent Laid-Open No. 4-335773, the density difference between the pixel of interest and the surrounding pixels is used as a pattern, and a pattern table created in advance is used as a reference to separate the character portion and the photograph portion.

In Japanese Patent Laid-Open No. 4-304776, gradation distribution and / or gradation change are calculated, and the type of image data is determined based on the calculated gradation distribution and / or gradation change. To do. In this way, fuzzy inference is performed based on the characteristic amount obtained from the change state of the density, and the suitability for the type of image such as halftone dot, photograph, and character is obtained, and the read image becomes halftone dot and character. Even if a photograph and the like are mixed, it is possible to judge them separately.

Further, when the character is not a black character on a white background but a white character on a black background, it is easier to read by black-and-white reversal when printing on paper or the like, and toner and ink are saved. become. However, when the image is reversed and printed, the image is not automatically discriminated, and the user needs to perform an operation such as intentionally switching to the reverse mode with a button or the like.

[0005]

As described above, in the conventional image processing apparatus, when the difference between the maximum value and the minimum value in the block is used for separating the character halftone dot and the photograph, the object in the photograph part is Areas such as the contours of which the density changes drastically are mistakenly recognized as character halftone dots. Similar erroneous recognition may occur even when the pattern table is used.

The present invention has been made in view of such a situation, and the character portion and the photograph portion are separated without erroneously recognizing a portion of the photograph portion where the density changes drastically, and the character portion is black on a white background. It is an object of the present invention to provide an image processing device which is capable of discriminating whether a character is white or black on a black background.

[0007]

In order to solve the above-mentioned problems, the present invention provides (1) a block black-and-white for obtaining a representative value of white pixels and a representative value of black pixels for each block of a multivalued image divided into blocks. The representative pixel value detecting means, the inter-block deviation integrating means for calculating the sum of the differences between the blocks of the representative values of the white pixels and the black pixels obtained by the block black and white representative pixel value detecting means, and the inter-block deviation integrating means And a character image identifying means for obtaining a feature amount indicating whether the block is a character or an image, based on the sum of the differences between the representative values of the white pixels and the black pixels between the obtained blocks, and (2) the character. It is characterized in that it has a character background identifying means for identifying whether a block determined to be a character by the image identifying means is a black character in a white background or a white character on a black background.

[0008]

According to the image processing apparatus of the present invention having the above-described structure, the representative value of the white pixels and the representative value of the black pixels are obtained for each block of the multi-valued image divided into blocks, and the obtained representative value of the white pixels and the black pixels. Calculate the sum of the differences between the blocks. A feature amount indicating whether the block is a character or an image is obtained from the sum of the differences between the representative values of white pixels and black pixels between blocks. In addition, the block determined to be a character is identified as a black character in a white background or a white character in a black background. In this way, the image is divided into blocks, the white pixel / black pixel representative value is obtained for each block, and the amount of change between blocks of the white pixel / black pixel representative value is checked to identify the character image and the photographic image. ,
By calculating the photographic degree, it is possible to identify whether the background is black or white in the character image.

[0009]

Embodiments will be described below with reference to the drawings. First, the discrimination between a character image and a photographic image will be described below with reference to FIGS. FIG. 1 is a diagram showing luminance and a histogram of an image of a fine character, FIG. 2 is a diagram showing luminance and a histogram of an image of a thick character, and FIG. 3 is a luminance and histogram of an image of a blank portion which does not include a character. FIG. 4 and FIG. 4 are graphs showing the black pixel representative value of each block of the document image written in black characters on a white paper, FIG.
Is a graph showing a white pixel representative value of each block of a document image written in black characters on a white paper, and FIG. 6 is a graph showing a black pixel representative value of each block of a photographic image, FIG. 7 is a graph showing a white pixel representative value of each block of a photographic image, and FIG. 8 is a graph showing a black pixel representative value of each block of a document image written in white characters on a blackboard. FIG. 9 and FIG. 9 are graphs showing the white pixel representative value of each block of the document image written in white characters on the blackboard.

When the pixels of the character image are divided into the background pixels and the character pixels, the thin line segment of the character is crushed, so that the brightness of the character pixel is not stable. FIG. 1 is a diagram showing a luminance histogram of an image of fine characters on a white paper surface. The horizontal axis represents luminance (256 gradations), and the vertical axis represents frequency. According to FIG. 1, the white pixels on the paper, which is the background, have a mountain-shaped distribution with a certain mass, whereas the distribution of black pixels of the character is
It has no clear mass. This occurs because the thin line segment of the character has been crushed. If the line segment does not collapse and the edge is sharp, the distribution of black pixels in the histogram has a more distinct cluster. Even if many small edges are blurred for a thick line segment, the central portion of the line segment can represent the original blackness of the character, and therefore the distribution of black pixels in the histogram has a clear cluster.

There are mainly two causes of the thin line segment being crushed. One is the resolution of the input device, and the other is the focus blur. For example, when inputting a paper surface to a scanner, an electronic still camera, or the like using an optical system using a CCD, if the resolution is low, the brightness will be averaged with the white of the background paper surface. Also, when there is no focus, when a pixel of a certain thin line segment is viewed, the light that should be input to the surrounding pixels will be input to the pixels of the line segment if the pixel is in focus. This is also because the brightness is averaged with white.

A character having a thin line segment is crushed for the above reason, and the brightness of the character pixel is not stable. In general, if the lighting conditions are the same, for example, the white brightness value of the background of the printed paper surface and the brightness value of the black ink of the character are the same values anywhere on the screen. Here, the image is divided into blocks of a certain size, and for example, the minimum luminance value of the pixels in the block is “character luminance” and the maximum luminance value is “background luminance”.
This value is obtained by the block black and white representative pixel value detection unit, which is defined as

Referring to FIG. 1, the brightness value of a character is b1.
(= 40), and the brightness value of the background is W1 (= 137).
The brightness of a character can take various values depending on the thickness of the line based on the above-described principle of line collapse. Therefore, the difference is likely to be large as compared with the brightness of the character in the adjacent block. On the other hand, the white paper part of the background has spaces between characters, spaces between lines, etc., so it is distributed as a group, so it will not be crushed like the character part and the brightness value of the adjacent block and background will be Even when compared, the difference is not as large as the luminance value of the character.

FIG. 2 is a diagram showing a luminance histogram in a block of the block adjacent to FIG. Since the block in FIG. 2 contains a thicker character than that in FIG. 1, the influence of the line collapse is not so small.
Compared to Fig. 1, the distribution of mountains in the letters is clearer,
The distribution of twin mountains is combined with the mountains in the background.

According to FIG. 2, the brightness value of the character is b2 (= 1
8), the brightness value of the background is W2 (= 139). Here, when the difference between the blocks of the brightness of the character is calculated, b2-
b1 = 18-40 = -22, and the difference in background brightness is w2-
w1 = 139-137 = 2. Comparing the absolute values of the differences, 22> 2, and the difference in the brightness of the character is larger. Furthermore, in a normal document image, characters do not always exist on the entire surface of the image and there are usually blank parts. However, when there are no characters in the block, the luminance histogram of the pixels in the block is Only the brightness value of the blank white background.

FIG. 3 is a diagram showing a histogram of a blank block containing no characters in the adjacent block of FIG.
In the case of a blank block, as shown in FIG. 3, a single mountain-shaped distribution is taken, and as shown in FIGS. 1 and 2, pixels are not distributed in the black portion (left side) of the mountain portion. When the brightness of the character of the block is forcibly obtained by the definition of the brightness of the character, a value close to the brightness value of the background cannot be avoided, and the value of b3 (= 11 in FIG.
It becomes the value of 9). The brightness value of the background is a value w3 (= 137) which is smaller than and larger than the brightness of the character. The blocks in Figure 3 are
Although taken from the blank part next to the block in FIG. 1, the difference in the background brightness value is w3-w1 = 137-137 = 0, and the difference in the character brightness value is b3-b1 = 119-40 = 79. Become. Comparing the absolute values of the differences, 79> 0, and the difference in the luminance of the character is definitely larger.

Mainly due to these two points, when the difference between the blocks of the luminance of the character and the luminance of the background is compared in absolute value, the difference in the luminance of the character becomes larger. In addition, since this comparison is performed between adjacent blocks, even if there are uneven brightness or shadows in the image, it can be said that if the blocks are sufficiently smaller than the brightness, unevenness, or shadows, they are viewed locally. The background brightness does not change significantly between blocks.

FIG. 4 shows a three-dimensional graph in which a certain document image is divided into blocks to obtain the luminance of a character, the luminance of the character in the height direction, and the vertical and horizontal axes represent the position of the block in the document. It is a figure. The data is the value on each grid point, but each grid point is connected by a straight line so that the state of change can be easily understood. The way to read the graphs is the same in FIGS. 5 to 9 described later. FIG. 5 is a diagram showing the same image as that of FIG. 4, showing a background luminance in the same manner in a block, the background luminance in the height direction, and a vertical axis / horizontal axis showing a three-dimensional graph showing the positions of blocks in a document. Is.

As can be seen at a glance by comparing FIGS. 4 and 5, the graph of the luminance of the characters shown in FIG. 4 is uneven, but the graph of the luminance of the background shown in FIG. It is gentle. In the graph of FIG. 5, especially in the peripheral part,
Although there is a sharply high portion, this is the portion of the blank block where there are no characters, indicating that there is a large difference from the block where the surrounding characters are present. Also,
In the graph of FIG. 5, the value on the vertical axis “Y block” is slightly depressed in the vicinity of 35, but this is caused by a shadow in this part during shooting, which makes it somewhat dark, and uneven brightness occurs. This is because However, as shown in FIG. 4, there is no significant difference from the surrounding blocks, and there is a smooth change, and even if there is uneven brightness, if the block is of an appropriate size, the change in background brightness between blocks is large. Indicates that it is not small.

Although the document image has been described above, a photographic image does not necessarily have a background of uniform brightness unlike the document image described above. Since a document image is an image that is usually printed on paper or the like, there was a uniform background called paper, but in a photographic image, various objects were photographed with various brightness distributions, and the paper of the document image There is no uniform background equivalent to.

FIG. 6 is a diagram showing a three-dimensional luminance graph of characters in a photographic image, and FIG. 7 is a diagram showing a three-dimensional luminance graph of the background of the photographic image. 6 and 7 do not have a gentle distribution as shown in FIG. Therefore, looking at the distributions of the brightness of the characters and the brightness of the background, it can be said that the document image has a smooth distribution, and the possibility of a photographic image is high if the change between blocks is large. In order to determine whether the change is drastic, the absolute value of the difference between the adjacent blocks of the luminance of the character and the luminance of the background is obtained for each block, and the sum is taken to quantify. The value of the sum is divided by the number of calculations between blocks and normalized so that it is not affected by the number of blocks. This gives a generalized value. This value is obtained by the inter-block deviation integration unit.

It can be said that when this value is large, the change is large, and when it is small, the change is small, but the threshold value is statistically determined. By setting not only one threshold value but two threshold values, it is possible to calculate which is closer to a document image or a photographic image, which is not clear. If the normalized inter-block change value is larger than a certain threshold value T1, it is determined that the change is severe and it is determined that the image is a photographic image, and the certain threshold value T1
If it is smaller than 2, it is judged as a document image, and if it is in between,
A comparison between the threshold values T2 and T1 makes it possible to quantitatively represent which and how close. Up to now, the explanation has been limited to the case where the document image is printed with black characters on white paper, but the brightness of the characters can be changed even for images such as blackboard with white chalk on which the characters are written. The principle is the same, assuming that is white and the background brightness is black.

FIG. 8 is a diagram showing a luminance (white) three-dimensional graph of characters on a blackboard, and FIG. 9 is a diagram showing a background luminance (black) three-dimensional graph. Comparing FIG. 8 and FIG. 9, it can be seen that the change in background brightness between blocks is smaller than the change in character brightness. Here, even in the case of a document image, the definition of the brightness of the character and the brightness of the background are changed to black and white depending on whether the printed matter is a black print on a white paper or a white print on the blackboard. But,
The definition is only a linguistic problem, and the actual work is to examine the change between the maximum brightness value and the minimum brightness value in the block between blocks, so it is important to know which is the background and which is the character. It is not necessary to set the brightness of the character, and the brightness of the character having a large change and the brightness of the background having a small change. With this, in the case of a document image, it is possible to determine whether the background is white or black.

If it is possible to determine whether the background is white or black, it is useful in printing when binarizing an image. For example, fax, scanner,
In copying machines and the like, when obtaining the threshold value, it is often practiced to set the threshold value to a luminance value that is slightly lower than the calculated threshold value, but if this is done, noise in the blank area will be reduced and the appearance will be reduced. Is very good. However, in the case of white letters on a black background, lowering the threshold value by the same method will result in increasing noise. Further, since the background is black, it is difficult to see, and a large amount of toner is consumed in a copier or the like. If it is known that the background is black by using the method of the present invention, the threshold value is reduced to a higher value to reduce noise, or black-and-white reversal is output at the time of printing for easy viewing, and toner consumption is low. The advantage of reducing the amount is born.

Therefore, the image is divided into blocks, and the value representing the white pixels and the black pixels in the block, for example, the maximum luminance value and the minimum luminance value between the blocks is observed, and one of them is small. It can be seen that there is a high possibility that it is a background document image, and if there is a large change in both, it is likely that it is a photographic image. The size of the area to be determined is the area of the block used when the absolute values of the changes in the background brightness and the character brightness between adjacent blocks are calculated and the sum thereof is calculated.

When the contrast between the maximum brightness value and the minimum brightness value in a block and the number of changes in the brightness value of an adjacent pixel are used as in the conventional case, a character image-like portion in a photographic image, for example, the outline of an object is used. May be erroneously recognized as a character part. Further, even in a document image, it is impossible to determine whether a blank portion having no character is a character area or an image area.
However, in the method of the present invention, since the change from the surrounding blocks is observed and the size of the contrast itself is not a problem, the character image-like portion in the photographic image, for example, the character portion in the outline of the object and the like. There is no false recognition.

Also, regarding the size of the region to be examined, there is a limit to the size of the block in the conventional method, and it is necessary to make a judgment for each fine block and correct the judgment result between the blocks. However, in the method of the present invention, since the change from the surrounding blocks is examined and then normalized, the increase or decrease in the number of blocks does not affect. Further, since the contents of the image are not recognized and judged like humans, the method of the present invention may cause misrecognition, but by setting two threshold values, even if it is difficult to clearly distinguish them. , It is possible to obtain the ratio of which is closer to the document or the photo. Therefore, when binarizing the photo part with error diffusion and the character part with a fixed threshold value, the error diffusion can be adjusted according to the ratio. It is possible to mix the fixed threshold value and the error diffusion according to the photographic degree.

Further, the brightness value of the character and the brightness value of the background are detected for distinguishing the document image and the photographic image, which is described in Japanese Patent Application No. 5-299609 previously proposed. There is also an advantage that it can be used to calculate a threshold value for binarization by any method. That is, a histogram of luminance values is obtained from a multi-valued image, a value that represents the white pixel value and the black pixel value in the image is obtained from the obtained histogram, and a binary value is obtained from the obtained white pixel representative value and black pixel representative value. The threshold for conversion is calculated. In addition, the threshold of binarization is calculated for each block of the multi-valued image divided into blocks, and it is detected whether or not a character or figure is included in the block. The threshold for binarization is calculated from the thresholds of surrounding blocks. The binarization threshold value obtained for each block is converted into a threshold value for each pixel, and the original multi-valued image is binarized using the obtained binarization threshold value for each pixel.

FIG. 10 is a block diagram for explaining one embodiment (claims 1 and 2) of the image processing apparatus according to the present invention. In the figure, 1 is a block black-and-white representative pixel value detection unit, and 2 is The block-to-block deviation integration unit 3 is a character image identification unit. The block monochrome representative pixel value detection unit 1 obtains a representative value of white pixels and a representative value of black pixels for each block of a multi-valued image divided into blocks. The inter-block deviation integration unit 2 calculates the sum of the differences between the blocks of the representative values of the white pixels and the black pixels obtained by the block monochrome representative pixel value detection unit 1. The character image identification unit 3 obtains a feature amount indicating whether the block is a character or an image, based on the sum of the differences between the representative values of the white pixels and the black pixels between the blocks obtained by the inter-block deviation integration unit 2. In addition, the block that is determined to be a character by the character image identification unit 3 has a character background identification unit that identifies whether the block is a black character in a white background or a white character on a black background.

The multi-grayscale luminance value a obtained from the original image is input to the block monochrome representative pixel value detection unit 1, and the block monochrome representative pixel value detection unit 1 inputs the white pixel representative value b and the black pixel representative value for each block. The value c is output. The block size may be fixed or may be adaptively changed depending on the image.
The blocks may be arranged side by side, or the blocks may overlap each other.

FIG. 11 is a diagram showing an example in which blocks are taken so as to overlap each other. 5 pixels vertically from a certain screen,
This is the part where five horizontal pixels are arranged.
For the sake of explanation, the vertical direction of pixels is called a row, and the horizontal direction is called a column. Rows are 1, 2, 3, 4, 5 from the top, and columns are a, b, c, d, and e from the left. When the block is divided into blocks each having a size of 3 pixels × 3 pixels, it is assumed that the adjacent blocks are overlapped by one row and one column.

Block p has columns c, d and e in rows 3, 4, and 5.
Occupy part of. As for the upper block q, the block p occupies the part of the columns c, d, and e in the rows 1, 2, and 3. Therefore, the pixels in the columns c, d, and e of row 3 overlap with the upper block q. In the left block q, the block p occupies the part of a, b, and c in the rows 3, 4, and 5. Therefore, the pixels in rows 3, 4, and 5 of column c overlap the left block r. under,
Repeat on the right as well. By taking the overlapping blocks in this way, it is possible to determine the document image and the photographic image in units smaller than the size of the block.

The white pixel representative value and the black pixel representative value are values representing the white pixel and the black pixel in the block, and are the same as those described in the above-mentioned binarization method. For simple calculation, for example, the maximum brightness value and the minimum brightness value in the block may be used, but if a good value that is not easily affected by noise is to be obtained, the one described in the binarization method described above is used. It is better to obtain it in the same way as.

12 and 13 are block diagrams of the block monochrome representative pixel value detecting unit. FIG. 12 shows a representative pixel value detecting unit for white pixels, and FIG. 13 shows a representative pixel value detecting unit for black pixels. . In FIG. 12, 11 is a white register, 12 is a decoder, 13-1 to 13-n are registers, 14 is a comparator,
Reference numeral 15 is an inverter, 16 is a white level counter, 17 is a comparator, and 18 is a selector.

Pixel value a sent from the original image of FIG.
Is sent to the selector 18, and the increment signal j is generated according to the value of the pixel value a. When 0, register 0 (10
Only the increment signal of 01) is generated and the pixel value a is 1
In the case of, only the increment signal of the register 1 (1002) is generated, and thereafter, the registers 255 are allotted up to the pixel value a of 255. This creates a histogram on the register. The initial value of the white register 11 is the maximum brightness value (255 in this case), and is given before the loop processing described below is started. The value after the loop processing is the representative value 206 of white pixels.

The beginning of the loop processing starts with the decoding of the value of the white register 11. The value of the white register 11 is decoded by the decoder 12, and registers 13-1 to 13 in which the histogram value of the brightness value of the white register 11 is stored.
The control signal f is determined so that any one of 3-n is selected. The histogram value g read out from the register selected only by the control signal f is stored in the comparator 14
Sent to and compared with 0. If the histogram value g is equal to 0, the decrement signal h becomes valid by the comparator 14 and the value of the white register 11 is decremented by 1. If it is not 0, the decrement signal i is validated by the comparator 14 and the inverter 15 and sent to the histogram registers 13-1 to 13-n and the white level counter 16.

The decrement signal i sent to the histogram register is valid only for the register selected by the control signal f, and the value of the selected register is decreased by one. The value of the white level counter 16 is sent to the comparator 17 and compared with 0. The value of the white level counter 16 is 0
If it is equal to, the representative value of white pixels will be obtained,
The value e of the white register 11 is output as a representative value of white pixels. If the value of the white level counter 16 is not 0, the process returns to the beginning again, and the loop process starting from the value of the white register 11 being decoded by the decoder 12 is repeated. The initial value of the white level counter 16 is a parameter and represents, when the pixels are arranged in the order of brightness, which brightness value of the brightest pixel is used as the representative brightness value of the white pixels. Generally, the value may be about several percent of the total number of pixels in the block.

FIG. 13 is a block diagram of a representative pixel value detection unit for black pixels. In the figure, 21 is a black register, 22 is a decoder, and 2 is a decoder.
3-1 to 23-n are registers, 24 is a comparator, 25 is an inverter, 26 is a black level counter, 27 is a comparator, and 28 is a selector. In FIG. 13, the histogram values are stored in registers 0 (23-1) to 255 (23-n). This value is the same as the histogram register in FIG. 2 or another register in which data is copied. When using the same histogram register, it is necessary to prevent access to the histogram register at the same time. The initial value of the black register 21 is the minimum brightness value (0 in this case), and is given before the loop processing described below is started. The value after the loop processing is the representative value e ′ of the black pixel.

The beginning of the loop processing starts with the decoding of the value of the black register 21. The value of the black register 21 is decoded by the decoder 22, and the registers 23-1 to 2-2 in which the histogram value of the brightness value of the black register 21 is stored.
The control signal f'is determined so that any one of 3-n is selected. The histogram value g'read from the register selected only by the control signal f'is sent to the comparator 24 and compared with 0. Histogram value g '
Is equal to 0, the decrement signal h'is enabled by the comparator 24 and the value of the black register 21 is decremented by 1. If not 0, comparator 24 and inverter 2
5, the decrement signal i'becomes valid, and the histogram registers 23-1 to 23-n and the black level counter 26 are activated.
Sent to.

The decrement signal i'transmitted to the histogram register is valid only for the register selected by the control signal f ', and the value of the selected register is decreased by one. The value of the black level counter 26 is the comparator 27
Sent to and compared with 0. If the value of the black level counter 26 is equal to 0, it means that the representative value of the black pixel has been obtained, and the value e ′ of the black register 21 is output as the representative value of the black pixel. If the value of the black level counter 26 is not 0, it returns to the beginning again, and the value of the black register 21 is set to the decoder 2
The loop process starting from being decoded by 2 is repeated.

The initial value of the black level counter 26 is a parameter, and represents, when the pixels are arranged in the order of luminance, the luminance value of the darkest pixel to be the representative luminance value of the black pixels. Generally, the value may be about several percent of the total number of pixels. The white pixel representative value b and the black pixel representative value c obtained from the block black-and-white representative pixel value detection unit 1 are stored in the inter-block deviation integration unit 2
The inter-block deviation integration unit 2 inputs the inter-block white pixel deviation integrated value d and the inter-block black pixel deviation integrated value e.
To get The absolute value of the difference between the white pixel representative value b of the target block and the white pixel representative value b of the surrounding blocks is calculated between a plurality of blocks in the region where the image type is to be identified, and the sum is calculated as the inter-block white pixel deviation integrated value. d. The inter-block black pixel deviation integrated value e is calculated in the same manner from the black pixel representative value c.

FIG. 14 is a diagram for explaining the calculation method of the white pixel deviation integration unit between blocks. Block vertically m
It is assumed that there are m and side by side m. For explanation, i from above
The block from the left and the jth block from the left is called a block (i, j), and the white pixel representative value of the block is w (i, j) and the black pixel representative value is b (i, j). Block ((i, j)
Blocks adjacent to are block (i-1, j), block (i + 1, j), block (i, j-1), block (i, j + 1)
It is four. The deviations of the white pixel representative values from these blocks are abs (w (i, j) -w (i-1, j)), abs (w (i, j)-
w (i + 1, j)), abs (w (i, j) -w (i, j-1)), abs (w
(i, j) -w (i, j + 1)). abs () means absolute value. This may be calculated between all adjacent blocks. The sum S is as follows.

[0043]

[Equation 1]

When divided by the number of blocks between blocks, num, it is normalized to a value per block, and the inter-block white pixel deviation integrated value d is obtained. num = (m-1) * (n-1) d = S / num By normalizing to a value per block, the number of blocks to be calculated can be increased, or the area of a block group can be formed only by a rectangle. Alternatively, the same threshold value can be used in the character image identification unit 3 even if the shape is various such as a circle. When the shape of the area of a group of blocks is other than a rectangle, the calculation method of the number num between blocks is different from the above expression, but it can take any shape as long as the number between blocks is counted. The black pixel deviation value e between blocks can be calculated similarly to the white pixel deviation d between blocks. The inter-block white pixel deviation integrated value d and the inter-block black pixel deviation integrated value e obtained from the inter-block deviation integrating section 2 are input to the character image identifying section 3 to obtain a feature amount f representing a character or an image.

FIG. 15 is an explanatory diagram of the character image identifying section. The horizontal axis represents the inter-block white / black pixel deviation integrated values d and e, which are larger toward the right and smaller toward the left. First, the smaller value g of the inter-block white pixel deviation integrated value d and the inter-block black pixel deviation integrated value e is obtained. As shown in the figure, if g is less than or equal to the threshold T1, the character portion
If it is above, it will be the photo department. When g is between the threshold value T1 and the threshold value T2, the photograph degree h is calculated by h = (g-T1) / (T2-T1).

The degree of photograph h is a value from 0 to 1, and the closer it is to 0, the higher the probability of being a character, and the closer it is to 1, the higher the probability of being a photograph. The feature amount f representing a character or an image is g
Is less than or equal to the threshold T1, f = 0, and more than or equal to the threshold T2 is f =
1, and if between the threshold values T1 and T2, f = h. In addition, the character portion or the area identified as the possibility of the character portion identifies whether the background of the character portion is a white background or a black background depending on which of d and e is smaller. If d is smaller, the background is white and the characters are black, and if e is smaller, the background is black and the characters are white.

[0047]

As is apparent from the above description, according to the present invention, the block black and white representative pixel value detecting means for obtaining the representative value of the white pixels and the representative value of the black pixels for each block of the multivalued image divided into blocks And an inter-block deviation accumulating means for calculating the sum of the differences between the blocks of the representative values of the white pixel and the black pixel obtained by the block black and white representative pixel value detecting means, and between the blocks obtained by the inter-block deviation integrating means White pixel,
Since the image processing unit has a character image identifying unit that obtains a feature amount indicating whether the block is a character or an image based on the sum of the differences between the representative values of the black pixels, the image is divided into blocks and white pixels / By determining the black pixel representative value and observing the amount of change between blocks of the white pixel / black pixel representative value, the character image and the photographic image are identified, the photographic degree is calculated, and the background of the character image is black or white. Can be identified.

[Brief description of drawings]

FIG. 1 is a diagram showing a luminance histogram of an image of a fine character according to the present invention.

FIG. 2 is a diagram showing a luminance histogram of an image of a thick character according to the present invention.

FIG. 3 is a diagram showing a luminance and a histogram of an image of a blank portion which does not include a character in the present invention.

FIG. 4 is a diagram showing a graph showing a black pixel representative value of each block of a document image written in black characters on a white paper according to the present invention.

FIG. 5 is a diagram showing a graph showing a white pixel representative value of each block of a document image written in black characters on a white paper according to the present invention.

FIG. 6 is a diagram showing a graph showing a black pixel representative value of each block of a photographic image in the present invention.

FIG. 7 is a diagram showing a graph showing a white pixel representative value of each block of a photographic image in the present invention.

FIG. 8 is a diagram showing a graph showing a black pixel representative value of each block of a document image written in white characters on a blackboard according to the present invention.

FIG. 9 is a diagram showing a graph showing a white pixel representative value of each block of a document image written in white characters on a blackboard according to the present invention.

FIG. 10 is a configuration diagram for explaining an embodiment of an image processing apparatus according to the present invention.

FIG. 11 is a diagram showing an example of a case where blocks are overlapped and taken in the present invention.

FIG. 12 is a configuration diagram of a representative pixel value detection unit for white pixels in the present invention.

FIG. 13 is a configuration diagram of a representative pixel value detection unit of a black pixel according to the present invention.

FIG. 14 is an explanatory diagram of an inter-block white pixel deviation integration unit according to the present invention.

FIG. 15 is an explanatory diagram of a character image identifying unit according to the present invention.

[Explanation of symbols]

1 ... Block black-and-white representative pixel value detection unit, 2 ... Inter-block deviation integration unit, 3 ... Character image identification unit.

Claims (2)

[Claims] 1. A block black and white representative pixel value detecting means for obtaining a representative value of white pixels and a representative value of black pixels for each block of a multi-valued image divided into blocks, and a white obtained by the block black and white representative pixel value detecting means. Inter-block deviation integrating means for calculating a sum of differences between blocks of representative values of pixels and black pixels, and a value of sum of differences of representative values of white pixels and black pixels between blocks obtained by the inter-block deviation integrating means Further, the image processing apparatus is provided with a character image identification unit that obtains a characteristic amount indicating whether the block is a character or an image. 2. A character background identification means for identifying whether the block determined as a character by the character image identification means is a black character in a white background or a white character on a black background. The image processing apparatus according to claim 1, which is characterized in that.