JP2000259765A - Character recognition device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve the recognition rate without requiring a long processing time and a large-scale memory by inverting and outputting the color of a pixel of interest in an input image when a pixel matrix having the pixel of interest in the center satisfies a specific expression. SOLUTION: A preprocess part 2 performs a filter process while scanning the input image and sends an output image to a segmentation part 3 when finishing the filter process for the whole image or a desirable specified area. The segmentation part 3 segments a character image of a parts where characters are prsent from the whole preprocessed image and sends it to a recognition part 4. The recognition part 4 recognizes the character image as characters and outputs the result in the form of character codes. Here, the filter used by the preprocess part 2 inverts and outputs the color of the pixel of interest in the input image when the pixel matrix X (i, j) of three rows and three columns having the pixel of interest in the center satisfies a conditional expression, etc. In the expression, 0 represents a white pixel, 1 represents a black pixel, and * is an arbitrary pixel whose color is not important.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

TECHNICAL FIELD The present invention relates to a facsimile (i, j)
And a character recognition device for recognizing an image in which distortion has been corrected after the image is read at a vertical / horizontal resolution ratio of 1: 2 as in the standard mode.

[0002]

2. Description of the Related Art In a standard mode usually used for FAX, a vertical resolution is approximately 100 dpi (dot per inc.).
h), the horizontal resolution is 200 dpi, and the vertical / horizontal resolution ratio is 1: 2. Therefore, if the obtained black and white pixels are arranged at an equal pitch, an image contracted in the vertical direction as shown in FIG. Since such an image is visually unnatural, it is necessary to compensate for the lack of information and output the image so that the image has an aspect ratio of 1: 1 before outputting the image. This correction is approximately performed by successively drawing each line twice. FIG. 7 shows an image obtained as a result of the correction.

On the other hand, since a character recognition algorithm has been developed in accordance with a standard character shape, it is not possible to recognize a contracted character as shown in FIG. Therefore, even when a FAX image is to be recognized, character recognition is performed on the corrected image as shown in FIG. As a method of character recognition,
Various algorithms have been devised. For example, for example, “Consideration on Dimensionality and Variable Conversion of Feature in Character Recognition” (Wakabayashi et al., IEICE Technical Report PRU92-115, 1993)
Statistical character recognition method using the feature amount of a figure, which is introduced elsewhere, is widely used because it is strong in character deformation and has an excellent recognition rate.

[0004] The direction index histogram is the basis of the character feature amount in this statistical character recognition method. FIG. 8 is an example of a direction index histogram in the aforementioned document (Wakabayashi et al.). In the example of FIG. 8, a contour image is first extracted from FIG. 8A, which is an image of a cut-out single character, to obtain FIG. 8B. It is divided into 7 × 7 small blocks.
As shown in (c), the connection state of the adjacent pixels of the contour image in the block is 0, 45, 90, 13 regardless of the forward or reverse direction.
The data is classified into five-degree four-direction components and tabulated, and a direction index histogram is obtained. FIG. 8D, which is the distribution of the direction index histogram obtained as a whole for each block in this manner, is the character feature amount. Not only the recognition method described in the above-mentioned document, but also in many other recognition methods, the distribution of connected components in the vertical, horizontal, and diagonal directions between adjacent pixels is used as a basic element of an important character feature amount.

[0005]

Looking at the FAX image in which the aspect ratio has been corrected in FIG. 7, most of the oblique components of the outline of the image are such that two black or white pixels continue in the vertical direction in the vertical direction, and 2
The pattern extends in the horizontal direction of the pixel. This is because each line of the original image is simply superimposed twice, so that all the parts that should have been oblique contour components have become vertical or horizontal components, resulting in a so-called jerky image. -ing Visually, the image of FIG. 7 lacks smoothness and is unnatural, but the problem is even more serious for character recognition. In this image, the distribution of the connected components of the contour of the original character image has been greatly distorted, so that the statistical character recognition method and many other recognition methods cannot be applied, and the connected components are not referred to. A recognition method with a poor recognition rate must be applied.

As a general smoothing method for smoothing the outline of an image, for example, there is a linear spatial filter that takes a weighted average of adjacent pixels. Since the small portion in the image is crushed, the recognition rate is rather lowered.
The present invention is to apply a statistical character recognition method which is generally well-established and has a high recognition rate to an image such as the standard mode of FAX (i, j). It is another object of the present invention to provide a pre-process for improving the distortion of a connected component of an outline of an image by a simple process.

[0007]

According to a first aspect of the present invention, there is provided a character recognition apparatus including at least an image input unit, a preprocessing unit for correcting an image, a cutout unit for cutting out a character image from an image, and a cutout character image. And a character recognition unit for recognizing pixels Pi, j (where 2 ≦ i ≦ M
−1, 2 ≦ j ≦ N−1), a three-row, three-column pixel matrix X (i, j), where 1 is a black pixel, 0 is a white pixel, and * is any pixel, If any one of Expressions 1, 2, 3, and 4 is satisfied, Pi, j = 1 is set.

[0008]

(Equation 1)

[0009]

(Equation 2)

[0010]

(Equation 3)

[0011]

(Equation 4)

According to a second aspect of the present invention, at least an image input unit, a preprocessing unit for correcting an image, a cutout unit for cutting out a character image from an image, and a character recognition unit for recognizing the cutout character image And the pre-processing unit comprises:
Pixel Pi, j (where 2 ≦ i ≦ M−1, 2 ≦ j ≦ N−
A 3 × 3 pixel matrix X (i, j) centered on 1) is
When 1 is a black pixel, 0 is a white pixel, and * is an arbitrary pixel,
If any of the following Expressions 5, 6, 7, and 8 is satisfied, Pi, j = 1 is set.

[0013]

(Equation 5)

[0014]

(Equation 6)

[0015]

(Equation 7)

[0016]

(Equation 8)

According to a third aspect of the present invention, at least an image input unit, a preprocessing unit for correcting an image, a cutout unit for cutting out a character image from an image, and a character recognition unit for recognizing the cutout character image are provided. And the pre-processing unit comprises:
Pixel Pi, j (where 2 ≦ i ≦ M−1, 2 ≦ j ≦ N−
A 3 × 3 pixel matrix X (i, j) centered on 1) is
When 1 is a black pixel, 0 is a white pixel, and * is an arbitrary pixel,
If either of the above equations 1 and 3 is satisfied, Pi, j =
It is characterized in that Pi, j = 0 if any one of Expressions 6 and 8 is satisfied.

According to a fourth aspect of the present invention, there is provided a character recognition apparatus including at least an image input unit, a preprocessing unit for correcting an image, a cutout unit for cutting out a character image from an image, and a character recognition unit for recognizing the cutout character image. And the pre-processing unit includes a pixel Pi, j (where 2 ≦ i ≦ M−1, 2 ≦ j ≦ N−
A 3 × 3 pixel matrix X (i, j) centered on 1) is
When 1 is a black pixel, 0 is a white pixel, and * is an arbitrary pixel,
If either of Equations 2 and 4 is satisfied, Pi, j =
It is characterized in that Pi, j = 0 if any one of Equations 5 and 7 is satisfied.

According to a fifth aspect of the present invention, there is provided a character recognition apparatus comprising: an image input unit; a preprocessing unit for correcting an image; a cutout unit for cutting out a character image from an image; and a character recognition unit for recognizing the cutout character image. And the pre-processing unit includes a pixel Pi,
j (however, 2 ≦ i ≦ M−1, 2 ≦ j ≦ N−1), a 3 × 3 pixel matrix X (i, j), where 1 is a black pixel, 0 is a white pixel, and * When any pixel is satisfied, Pi, j = 1 is satisfied if any of Expressions 1 and 4 is satisfied.
If any of the above equations 6 and 7 is satisfied, Pi,
It is characterized in that j = 0.

According to a sixth aspect of the present invention, there is provided a character recognition apparatus for recognizing a binarized image composed of M rows and N columns of white or black pixels in which the difference in the vertical and horizontal resolution ratios has been corrected. , At least an image input unit, a preprocessing unit that corrects the image, a cutout unit that cuts out a character image from the image, and a character recognition unit that recognizes the cutout character image, wherein the preprocessing unit includes a pixel Pi , J (where 2 ≦ i ≦ M−1,
3 ≦ 3 pixel matrix X centered on 2 ≦ j ≦ N−1)
When (i, j) is 1 for a black pixel, 0 for a white pixel, and * for any pixel, if either of the above equations 2 and 3 is satisfied, then Pi, j = 1, and If any of Expressions 5 and 8 is satisfied, Pi, j = 0 is set.

As described above, in the character recognition device according to the first to sixth aspects of the present invention, as a result of correcting the difference in the vertical and horizontal resolution ratios, the original oblique component of the contour becomes a vertical or horizontal component. Such a binarized image is subjected to a smoothing process appropriate for a known character recognition method, and the oblique component of the contour is reproduced again, thereby making it possible to greatly improve the recognition rate. The content of the smoothing process differs for each invention described in each claim.

[0022]

FIG. 1 shows a processing configuration of a character recognition device according to the present invention. The image input unit 1 inputs an image read from an external device by facsimile or the like. When the vertical and horizontal resolutions are different in the FAX standard mode, an image in which the distortion of the aspect ratio is corrected by the method described in the section of the related art is input and passed to the preprocessing unit 2. The preprocessing unit 2 performs a filtering process described below while scanning the input image, and ends the filtering process on the entire image or a designated desired region.
The output image is sent to the cutout unit 3.

The cutout unit 3 cuts out a character image where a character exists from the whole image that has been subjected to the preprocessing, and sends it to the recognition unit 4. The recognition unit 4 recognizes the character image as a character and outputs the result as a character code. The text output unit has a function of outputting a recognition result from a device to the outside in a format according to a request, and a unit corresponding to various code sets and interfaces is used.

The pre-processing unit 2 does not necessarily need to be before the clipping unit 3, but it is simpler to perform a filtering process before the image is finely clipped. However, considering the time required for the filtering process, it is shorter in terms of time to perform the filtering process only on the image after the clipping. Therefore, in FIG. 1, the order of the preprocessing unit 2 and the clipping unit 3 is changed. Is also good. In the following embodiment, for the sake of simplicity of understanding, an image cut out into one character first is targeted.
The case where the preprocessing is performed will be described. Even if the clipping is performed after the pre-processing, there is no essential difference only by performing the pre-processing on all the character images at once. Conventional preprocessing such as removal of noise included in an input image may be performed by the image input unit 1 or the preprocessing unit 2.

Hereinafter, a filter used in the preprocessing unit 2 of the present invention will be described. In the equations used in the following description, 0 represents a white pixel, 1 represents a black pixel, and * represents any pixel color. In the filter used in the present invention, the pixel matrix X (i, j) of 3 rows and 3 columns centering on the pixel of interest in the input image is obtained by using the above formulas 1, 2, 3, 4,
When a specific conditional expression among Expressions 5, 6, 7, and 8 is satisfied, the color of the pixel of interest is inverted and output.

FIG. 2 is a diagram showing the operation of each filter used in the present invention in the arrangement of white pixels and black pixels. Also in the figure, * indicates that the color of the pixel is not considered. In FIG.
a1, a2, a3, a4, b1, b2, b3, b4 each describe the operation of the filter with that name. Pixel matrix X (i, j) of 3 rows and 3 columns obtained from the input image
, The target pixel Pi, j at the center is inverted. That is, the pixel array on the left side of the arrow corresponds to the conditional expression used by each filter, and the fact that the input pixel matrix matches the pixel array on the left side is nothing but satisfying the conditional expression. When the input pixel matrix X (i, j) of 3 rows and 3 columns matches the left array, the pixel of interest Pi, j at its center is inverted, and as a result, the pixel matrix X
(I, j) changes to the array shown to the right of the arrow.

The filters a1, a2, a3, and a4 are filters using Equations 1, 2, 3, and 4 as conditional expressions, respectively. In any of the filters using conditional expressions 1 to 4, there is a possibility that the pixel matrix X (i, j) of the input image satisfies the condition only when the target pixel is white.
When the condition is satisfied, the target pixel is inverted from white to black.
These four filters a1 to a4 are collectively referred to as a group a. For example, the filter a1 is such that when the center pixel of interest is a white pixel, the upper right, right, lower, and lower left pixels of the pixel of interest are black pixels, and the upper and left pixels of the pixel of interest are white pixels,
This is a filter that converts a target pixel into a black pixel and outputs the converted pixel.

On the other hand, filters b1, b2, b
Reference numerals 3 and b4 denote filters using Expression 5, Expression 6, Expression 7, and Expression 8, respectively, as conditional expressions. These filters using any one of Expression 5, Expression 6, Expression 7, and Expression 8 as the conditional expressions only allow the pixel matrix X (i, j) of the input image to be a conditional expression if the target pixel is black. May be satisfied, and if so, the pixel of interest is converted to white and output. These four filters b1 to b4 are collectively referred to as a group b. For example, in the filter b1, the center pixel of interest is a black pixel,
The upper right, upper, left and lower left pixels of the target pixel are white pixels,
When the right pixel under the target pixel is a black pixel, the filter converts the target pixel into a white pixel and outputs the converted white pixel.

In summary, when the input pixel matrix X (i, j) satisfies the expression 1, the filter a1 belonging to the group a inverts the target pixel at the center of the matrix from white to black, and the filter a2 expresses the expression 2
Is satisfied, the target pixel is similarly inverted to black. Similarly, the filter a3 inverts the target pixel to black when the expression 3 is satisfied, and the filter a4 inverts the target pixel to black when the expression 4 is satisfied. Things. Filter b1 belonging to group b
Inverts the center pixel of interest from black to white when X (i, j) satisfies Equation 5, and so on, filter b2 inverts to white when Equation 6 is satisfied, and filter b3 calculates Equation 7 The filter b4 is inverted to white when satisfied, and the filter b4 is inverted to white when Expression 8 is satisfied.

In short, these filters invert the target pixel at the center of the matrix if the pixel matrix X (i, j) of three rows and three columns centering on the target pixel satisfies the conditional expression, If the conditional expression is not satisfied, there is an effect that the target pixel is output as it is without being inverted. When the target pixel is located at the end row or column of the input image, there is no pixel matrix X (i, j) of three rows and three columns centering on the target pixel, and therefore, is not subjected to the filtering process. However, if the character image is completely inside the input image, such a situation cannot occur, and conversely, if a part of the character protrudes from the input image and the character is missing, even if filter processing is performed There are almost no practical problems because they cannot be recognized in most cases.

A1, a2, a3 and a4 of group a
Each of the three filters converts a white pixel into a black pixel, and a1 processes an outline corresponding to the upper side of the image from the upper right to the lower left among the outlines of the image. A contour line corresponding to the lower side of the image going to the lower left, a3 is a contour line for the upper side going from upper left to lower right, and a4 is a contour line for the lower side going from upper left to lower right. B1, b2, b belonging to group b
Each of the four filters 3 and b4 converts a black pixel into a white pixel, and at the same time, similarly to the group a, b1 processes the contour of the upper edge of the image from the upper right to the lower left among the contours of the image. In the same manner, b2 processes the contour of the lower side of the image from upper right to lower left, b3 processes the contour of the upper side from upper left to lower right, and b4 processes the contour of the lower side from upper left to lower right. is there.

Here, when the roles of the filters of the groups a and b are arranged, the group a converts white pixels into black pixels,
The b group converts black pixels into white pixels, but at the same time, a
In each of the groups b and b, the first filter acts on the contour of the upper side of the diagonally shaded image from upper right to lower left, and the second filter acts on the contour of the lower side of the diagonally shaded image in the same direction. The third filter acts on the contour of the upper side of the oblique line from the upper left to the lower right, and the fourth filter acts on the contour of the lower side of the oblique line in the same direction. This can be easily understood by comparing the arrangement of the white pixels and the black pixels of each filter in FIG. 2 with, for example, the outline of the oblique line in the input image in FIG.

Considering the role of each filter, basically, if the combination is such that the upper side contour and the lower side contour are processed in each of the upper right and lower left directions and the upper left and lower right directions, the groups a and b It can be seen that any combination between groups is possible. For the four roles, a
Since a filter is selected from the group or the group b, it can be seen that all combinations have 16 powers of 2 4. Table 1 shows a list of 16 filter sets (filter combinations).

[0034]

[Table 1]

In the preprocessing unit of the present invention, a filter set of a preferable combination among 16 combinations of filters shown in Table 1 is applied to a pixel matrix of 3 rows and 3 columns centering on a target pixel in an input image. Is applied to attempt the conversion of the pixel of interest, and the processing is sequentially repeated until the processing on all the pixels except for the pixels at the end row or column of the image in which a three-row, three-column pixel matrix cannot be formed. Hereinafter, selection of this filter combination will be described. Among the combinations in Table 1, a filter set S1 using only a filter of group a is a pixel matrix X
When (i, j) satisfies any of the conditions of Expressions 1, 2, 3, and 4, the target pixel Pi, j is inverted from white to black.

FIG. 3 shows an output image obtained by applying S1 in Table 1, that is, a filter set including filters a1, a2, a3, and a4 to the corrected input image of FIG. FIG.
The original diagonal connected component is reproduced in the outline of the character image, and it can be expected that a good recognition result can be obtained even in character recognition. A1, a2, a3 belonging to the filter set S1,
Since all the filters in a group a4 convert white pixels into black pixels, comparing FIG. 3 which is an output image obtained using the filter set S1 with FIG. 7 which is the original image, The oblique line in FIG. 3 is thicker than the original image.

A filter set S2 using only the b group of filters in combination has a pixel matrix X (i, j) expressed by the following equation (5).
When any of the conditions of Expressions 6, 7, and 8 is satisfied, the target pixel Pi, j is inverted from black to white. The correction input image shown in FIG. 7 includes S2 in Table 1, that is, b1, b2, b3, and b4.
FIG. 4 shows an output image obtained by applying the filter set including the filters (1) to (3). FIG. 4 is also similar to FIG.
The oblique connected components are well reproduced. Filter set S
B1, b2, b3, and b4 of the b group constituting filter 2 convert a black pixel into a white pixel.
The oblique line of the image of FIG. 4 obtained using this filter set S2 is thinner than the original image.

Thickness and thinness of the line are not so affected by the recognition method, but there are also recognition methods affected by the thickness of the line. Therefore, it is necessary to devise a filter that does not change the thickness of the line so much. In order to maintain the thickness of the oblique line in the original image, a filter belonging to group a,
It is necessary to appropriately combine filters belonging to group b. For example, a filter of group a is selected so as to convert white pixels into black pixels in the upper contour of an oblique line image from upper right to lower left, and black pixels are converted into white pixels in the lower contour of the same oblique line image. If the filter of the b group is selected as described above, the thickness of the diagonal line does not change. Similarly, with respect to the image of the diagonal line from the upper left to the lower right, if the pair of the filters of the group a and the group b is selected, the original thickness of the diagonal lines in both directions can be maintained, so that the thickness of the horizontal line and the vertical line Without losing balance,
Character image features are saved.

In Table 1, the four types of filter sets S3 to S6 always select one filter from the groups a and b on the upper side and the lower side for oblique lines in either direction. The width does not change. Filter set S3
Is composed of a combination of filters a1, b2, a3 and b4. That is, if the pixel matrix X (i, j) satisfies the expression 1 or 3, the central pixel of interest is inverted from white to black, and if the expression 6 or 8 is satisfied, the central pixel of interest is inverted from black to white. I do.

In S3, a1 and b2 are paired, and among the contours in the upper right and lower left directions, a
Since the number of black pixels increases by the action of 1 and the number of white pixels increases by the action of b2 in the outline of the lower side of the image, the line width of the oblique line in the upper right and lower left directions does not change as a whole, and the center of gravity of the oblique line in this direction moves upward. The filters a3 and b4 are also paired,
Of the contours in the upper left and lower right directions, the number of black pixels increases by the action of a3 in the upper side contour and the number of white pixels increases by the action of b4 in the lower side contour, so that the line width of the oblique lines in the upper left and lower right directions does not change. FIG. 5 shows an example of an image processed using the filter set S3. Compared to FIG. 3 or FIG. 4, it can be seen that the original hatched width is maintained in FIG.

The filter set S4 includes b1, a2, b3,
a4 filter, and the pixel matrix X (i, j) is expressed by Equation 2
Alternatively, if Expression 4 is satisfied, the pixel of interest is inverted from white to black,
If Expression 5 or Expression 7 is satisfied, the target pixel is inverted from black to white. That is, in any of the upper right direction and the lower left direction and the upper left direction and the lower right direction, the number of white pixels increases in the outline of the upper side and the number of black pixels increases in the outline of the lower side, so that the line width of the oblique line does not change.

The filter set S5 includes a1, b2, b3,
a4, and the pixel matrix X (i, j) is expressed by the following equation 1.
Alternatively, if Expression 4 is satisfied, the pixel of interest is inverted from white to black,
If Expression 6 or Expression 7 is satisfied, the target pixel is inverted from black to white. That is, in the upper right-lower left direction, the number of black pixels increases in the outline of the upper side and the number of white pixels increases in the outline of the lower side.
Conversely, in the direction from the upper left to the lower right, white pixels increase in the outline of the upper side and black pixels increase in the outline of the lower side. In any case, the line width of the oblique line does not change.

The filter set S6 includes b1, a2, a3,
b4, and the pixel matrix X (i, j) is expressed by the following equation 2.
Alternatively, if Expression 3 is satisfied, the pixel of interest is inverted from white to black,
If Expression 5 or Expression 8 is satisfied, the target pixel is inverted from black to white. That is, in the upper right-lower left direction, white pixels increase in the outline of the upper side and black pixels increase in the outline of the lower side.
Conversely, in the direction from the upper left to the lower right, the number of black pixels increases in the outline of the upper side, and the number of white pixels increases in the outline of the lower side. This is a symmetrical operation with respect to the filter set S5, but the line width of the diagonal lines does not change in any direction in the filter set 6 as well.

As can be inferred from the above, in the filter set from S3 to S6, the line width of the diagonal line is not preserved, but the center of gravity of the diagonal line is moved in the up, down, left, or right direction by 1/2 pixel. I do. This is because the number of black pixels on one side of the diagonal line increases and the number of black pixels on the opposite side decreases as a result of the filter processing. Depending on which black pixel increases in each diagonal line in two directions, the upper part in S3 and the lower part in S4. , S5, and the center of gravity of the character image moves rightward in S6.

When the remaining sets after S7 are used, for example, the upper left and lower right diagonal lines become thicker and the upper right and lower left diagonal lines become thin, resulting in an imbalance in line width. However, a diagonal line smoothing effect is obtained. Therefore, it is also possible to employ these filter sets. S7 and S8 are combinations in which one diagonal line is thick and the other diagonal line is thin.
S9 to S12 are combinations in which one oblique line is thick and the other is unchanged in line width. In addition, S13 to S16 are combinations in which one oblique line is thin and the other line width does not change.

[0046]

According to the character recognition device of the present invention, the outline of the diagonal lines required for character recognition can be smoothed by simple filter processing, so that processing time is not required and a large-scale memory is not required. The recognition rate can be improved. Also, since smoothing processing is applied only to the parts where the outline is jagged with diagonal lines instead of smoothing the whole uniformly, character collapse does not occur and even if applied to images in fine mode, there is no adverse effect. Do not give. Furthermore, since the character recognition device according to the first aspect smoothes while increasing the line width of the oblique lines, if the image recognition apparatus is applied to a faint image, the effect of improving the line width is also exhibited.

According to the character recognition device of the second aspect, since the smoothing is performed while reducing the line width of the diagonal lines, while the line width of the diagonal lines is widened in the first aspect, the character recognition apparatus can be applied to an image that tends to be too dark. For example, the line width can be improved. Claim 3
Since the character recognition devices of Nos. 1 to 6 perform smoothing without changing the line width of oblique lines, they are suitable for a character recognition method in which the line width is important as information.

[0048]

[Brief description of the drawings]

FIG. 1 is a diagram showing a processing configuration of a character recognition device of the present invention.

FIG. 2 is a diagram showing the operation of the filter of the present invention.

FIG. 3 is a diagram showing an output image when a filter set S1 of the present invention is used.

FIG. 4 is a diagram showing an output image when a filter set S2 of the present invention is used.

[0049]

FIG. 5 is a diagram showing an output image when a filter set S3 of the present invention is used.

FIG. 6 is a diagram showing an image input in a vertical direction at a resolution of 横 of a horizontal direction.

7 is a diagram showing an image in which the aspect ratio of the image in FIG. 6 is improved.

FIG. 8 is a diagram showing how to obtain a direction index histogram in the character recognition method.

[0050]

[Explanation of symbols]

 1: Image input unit 2: Preprocessing unit 3: Cutout unit 4: Recognition unit 5: Text output unit

 ──────────────────────────────────────────────────続 き Continued on the front page F-term (reference)

Claims (6)

[Claims] 1. A character recognition device for recognizing a binarized image composed of white or black pixels in M rows and N columns in which the difference in the vertical and horizontal resolution ratios has been corrected, comprising at least an image input unit and an image input unit. It has a pre-processing unit for correction, a cut-out unit for cutting out a character image from an image, and a character recognition unit for recognizing the cut-out character image. The pre-processing unit includes pixels Pi, j (where
3 rows 3 centered on 2 ≦ i ≦ M−1, 2 ≦ j ≦ N−1)
When the pixel matrix X (i, j) of the column is 1 as a black pixel, 0 as a white pixel, and * as an arbitrary pixel, any one of the following formulas 1, 2, 3, and 4 is satisfied. For example, Pi, j = 1. (Equation 1) (Equation 2) (Equation 3) (Equation 4) 2. A character recognition apparatus for recognizing a binarized image composed of M rows and N columns of white or black pixels in which a difference in the resolution ratio between the vertical and horizontal directions has been corrected. It has a pre-processing unit for correction, a cut-out unit for cutting out a character image from an image, and a character recognition unit for recognizing the cut-out character image. The pre-processing unit includes pixels Pi, j (where
3 rows 3 centered on 2 ≦ i ≦ M−1, 2 ≦ j ≦ N−1)
When the pixel matrix X (i, j) of the column is 1 as a black pixel, 0 as a white pixel, and * as an arbitrary pixel, any of the following formulas 5, 6, 7, and 8 is satisfied. For example, Pi, j = 1. (Equation 5) (Equation 6) (Equation 7) (Equation 8) 3. A character recognition device for recognizing a binarized image composed of M rows and N columns of white or black pixels in which a difference in resolution ratio between height and width is corrected, comprising at least an image input unit and an image input unit. It has a pre-processing unit for correction, a cut-out unit for cutting out a character image from an image, and a character recognition unit for recognizing the cut-out character image. The pre-processing unit includes pixels Pi, j (where
3 rows 3 centered on 2 ≦ i ≦ M−1, 2 ≦ j ≦ N−1)
When 1 is a black pixel, 0 is a white pixel, and * is an arbitrary pixel, X, (i, j) of a column, if either of the above equations 1 and 3 is satisfied, Pi, j = The character recognition device is characterized in that Pi, j = 0 if the above expression is 1, and if any of the above expressions 6 and 8 is satisfied. 4. A character recognition device for recognizing a binarized image composed of M rows and N columns of white or black pixels in which the difference between the vertical and horizontal resolution ratios has been corrected, comprising: an image input unit; It has a pre-processing unit for correction, a cut-out unit for cutting out a character image from an image, and a character recognition unit for recognizing the cut-out character image.
≦ i ≦ M−1, 2 ≦ j ≦ N−1), a 3 × 3 pixel matrix X (i, j), where 1 is a black pixel, 0 is a white pixel,
* Represents an arbitrary pixel, and if either of the above equations 2 and 4 is satisfied, Pi, j = 1, and the above equation 5,
A character recognition device characterized by setting Pi, j = 0 if any one of Expressions 7 is satisfied. 5. A character recognition apparatus for recognizing a binarized image composed of M rows and N columns of white or black pixels in which a difference in the resolution ratio between the height and width is corrected, comprising: an image input unit; And a character recognizing unit for recognizing the cut-out character image. The pre-processing unit includes a pixel Pi, j (where 2 ≦ i ≦ M
−1, 2 ≦ j ≦ N−1), a three-row, three-column pixel matrix X (i, j), where 1 is a black pixel, 0 is a white pixel, and * is an arbitrary pixel, A character characterized by setting Pi, j = 1 if any of Expressions 1 and 4 are satisfied, and setting Pi, j = 0 if any of Expressions 6 and 7 are satisfied. Recognition device. 6. A character recognition device for recognizing a binarized image composed of M rows and N columns of white or black pixels in which the difference between the vertical and horizontal resolution ratios has been corrected. It has a pre-processing unit for correction, a cut-out unit for cutting out a character image from an image, and a character recognition unit for recognizing the cut-out character image. The pre-processing unit includes pixels Pi, j (where
3 rows 3 centered on 2 ≦ i ≦ M−1, 2 ≦ j ≦ N−1)
When the pixel matrix X (i, j) of a column is 1 for a black pixel, 0 for a white pixel, and * for an arbitrary pixel, if either of the above equations 2 and 3 is satisfied, Pi, j = 1; and if any of Equations 5 and 8 is satisfied, Pi, j = 0.