JPH0746371B2 - Character reader - Google Patents

Description

TECHNICAL FIELD The present invention relates to a character reading device, and in particular,
The present invention relates to a character reading device for reading low-quality engraved characters, printed characters, handwritten characters, etc., which are imaged with an industrial camera or the like, and which are missing or hidden.

[Conventional technology]

A character reading device that reads characters in an image can be realized by combining a character pattern cutout device that cuts out individual characters from an image and a character recognition device that recognizes the cut out character pattern.

FIG. 3 shows, for example, a conventional character pattern cutting device disclosed in Japanese Patent Publication No. 55-42434 and Japanese Patent Publication No. 53-46062.
It is a block diagram of the conventional character reading device which combined with the conventional character recognition device shown in the publication.

In the figure, 1 is an image memory for storing a digital image such as an input form, and 2 is a character frame set on a medium such as a form or known character arrangement information such as the position and size of characters to be printed or stamped. Stored character array information storage memory,
Reference numeral 3 denotes a character cutting means for cutting out a character pattern from the image memory 1 by referring to the contents of the character array information storage memory 2.
4 is a dictionary memory that stores a recognition dictionary that stores reference patterns corresponding to categories to be recognized, and 5 is an image memory 1.
It is a character recognition means for determining the category of the input character pattern by superimposing the character pattern cut out from the reference pattern of each category stored in the dictionary memory 4. The character cutout means 3 and the character recognition means 5 are realized by a microprocessor or the like.

FIGS. 4 (a) to 4 (i) are diagrams showing an operation example of the conventional character reading device shown in FIG.

6A shown in FIG. 3A is an input image stored in the image memory 1, 6a is a character pattern of category "3", 6b is a character pattern of category "4", 6c is a character pattern of category "8", 6d is a character pattern of category "7", and 7 shown in FIG. 7 (b) is a rectangular representation of the position and size of the character which is the contents of the character arrangement information storage memory 2, and 7a, 7b, 7c. , 7d are the 1st, 2nd and 3rd characters respectively from the left.
A rectangle representing the position and size of the fourth character and the fourth character, and 8 shown in (c) of the figure is a table in which the contents of the character array information storage memory 2 are represented by numerical values, 8a, 8b, 8c, 8d. Is the position number (NO) of the rectangle 7a, 7b, 7c, 7d, and X at the upper left corner, respectively.
The coordinates (SX), the Y coordinate (SY) at the upper left corner, the width (W), and the height (H) are expressed by numerical values. On the other hand, 9 shown in (d) of the figure is a peripheral distribution created from the input image 6, 9a is a peripheral distribution in the Y direction, 9b is a peripheral distribution in the X direction, and 10 shown in (e) is the input image. Character pattern rectangles cut out from 6, 10a, 10b, 10c, 10d are rectangles 7a, 7 respectively.
The rectangle of the character pattern cut out corresponding to b, 7c, 7d,
Reference numeral 11 shown in FIG. 6F is a recognition dictionary stored in the dictionary memory 4, and reference numerals 11a to 11j are reference patterns of categories "0" to "9", respectively. Further, 12 shown in FIG. 7G is a character pattern of the input image 6 corresponding to the rectangle 10a of the character pattern, and 13 shown in FIG. 7H is a character pattern 12 and the recognition dictionary 11 stored in the dictionary memory 4. Each category “0” to “9”
The candidate categories up to the fifth rank obtained by superimposing the reference patterns 11a to 11j of
13e is the similarity when the reference patterns 11d, 11i, 11j, 11a, 11f and the character pattern 12 are overlapped, and 14 shown in FIG. It is the recognition category “3” determined by the above.

Next, the operation will be described. The input image 6 is stored in the image memory 1.

The character clipping means 3 scans the input image 6 in the X and Y directions, counts black pixels for each column and each line, and creates a peripheral distribution 9b in the X direction and a peripheral distribution 9a in the Y direction. Next, the marginal distributions 9a and 9b are each binarized, and a rectangle 10 to be externally attached to the character pattern from the X coordinates of both ends of the black portion continuous in the X direction and the Y coordinates of both ends of the black portion continuous in the Y direction.
The positions and sizes of a, 10b, 10c, and 10d are calculated, and compared with the numerical values (position and size) of the table 8 which is the contents of the character array information storage memory 2 set in advance (the ratio is calculated. ) Rectangle 7a, 7b, 7c, 7d in the character pattern rectangle 10a, 10b, 10
The character pattern rectangle 10 is determined by associating c and 10d with each other, and the character pattern corresponding to the character pattern rectangle 10 is cut out from the input image 6. For example, the character pattern 12 is the input image 6 corresponding to the rectangle 10a of the character pattern.
Pattern.

Next, in the character recognition means 5, the character pattern 12 cut out by the character cutting means 3 and the recognition dictionary stored in the dictionary memory 4 are used.
Reference pattern 11a of the recognition target category stored in 11
~ 11j are overlapped with each other and the degree of matching (similarity) between them is obtained.

In this case, for example, for input character patterns and reference character patterns, a binary matrix in which white pixels are −1 and black pixels are +1 Then, the similarity is represented by S shown below.

Here, the numerator is a value obtained by subtracting the number of mismatches from the number of matches of the elements of the two matrices, and the denominator N is the total number of elements of the matrix. still, Is a vector and P and Q are scalars.

For example, the similarities 13a, 13b, 13c, 13d, 13e are the values of the similarity S calculated by the above formula when the reference character patterns 11d, 11i, 11j, 11a, 11f and the character pattern 12 are superposed, respectively. .

Next, a category having a high degree of similarity among all the recognition target categories “0” to “9” is set as a candidate category, and a category satisfying a condition defined by the recognition method from the candidate category and the degree of similarity 13, for example, the degree of similarity The highest category "3" is determined as the recognition category 14.

[Problems to be solved by the invention]

Since the conventional character reading device is configured as described above, if a character is missing or partially obscured by an obstacle depending on the direction and position of the image taken by an industrial camera, etc., the character can be accurately cut out. There was a problem that it did not exist.
Further, even when the character is cut out, there is a problem that it is difficult to correctly recognize the character due to lack or occlusion.

The present invention has been made in order to solve the above problems, and provides a character reading device that can accurately cut out a character and correctly recognize the character even if a part of the character is missing or hidden. With the goal.

[Means for solving problems]

The character reading device according to the present invention preferentially cuts out a character pattern which is obviously a character based on the value of a pixel of an input image, a priority character cutting means, and a character cut out by known character arrangement information and priority character cutting means. The position and size of the remaining characters are estimated based on the position and size information of the pattern to cut out the character pattern, and the estimated character cutting means and the character is missing based on the value of the pixel in the area formed by the character pattern, or Character recognition is performed by comparing the above-mentioned character pattern with the reference pattern of the character to be recognized in the area except for the missing hidden area detected by the missing hidden area detecting means and the missing hidden area detecting means for detecting the hidden area. And a character recognizing means for doing so.

[Action]

The character reading device according to the present invention is a character in which a character is missing or hidden, which is estimated from character arrangement information such as a predicted position and size of a character in advance and information of a character pattern already cut out by the priority character cutting means. Is cut out by the estimated character cutting-out means and the area of the lacking occlusion is detected, so that the character recognizing means recognizes the character using the pattern excluding the lacking occlusion area. Therefore, it is possible to accurately cut out a character even when the character is missing or hidden due to a change in the imaging direction or the position. Further, even if the character is missing or hidden, the character can be accurately recognized.

〔Example〕

 An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram of a character reading device according to this embodiment.
In the figure, 1 is an image memory for storing a digital image obtained by capturing an image of a medium on which a character to be recognized is recorded with an image capturing device such as a television camera, and 3a is an X direction and Y obtained by scanning the image memory 1. Priority character cutting means for preferentially cutting out a character pattern that is clearly a character based on the peripheral distribution value in the direction, 3b is the position and size of the character pattern already cut out of the character pattern that could not be cut out by the priority cutting means 3a. Estimated character cut-out means for cutting out by estimating from the size information and the contents of the character array information storage memory 2, 15
Is a missing hidden area detecting means for detecting an area where a character is missing or hidden based on a peripheral distribution value in the Y direction obtained by scanning an area formed by the character pattern cut out by the estimated character cutting means 3b, and 5a is a missing hidden area. It is a character recognition means for recognizing a character by comparing a character pattern with a reference pattern of a character to be recognized in an area excluding the area detected by the area detection means 15. The respective means 3a, 3b, 15, 5a are realized by a microprocessor or the like, and the memories 2 and 4 are the same as the conventional character reading device shown in FIG.

FIGS. 2A to 2N are diagrams showing an operation example of the character reading device according to the embodiment shown in FIG.

Reference numeral 16 shown in FIG. 3A is an input image stored in the image memory 1, and 16a to 16h are categories "3", "4",
The character patterns are "8", "7", "6", "2", "3", and "4". The character patterns 16a and 16b hide the upper part of the characters, and the character patterns 16g and 16h are characters. Is missing the top of the. Reference numeral 17 shown in FIG. 2B represents the position and size of the character, which is the content of the character array information storage memory 2, in a rectangle, and 17a to 17d are 1 to 4 characters from the left of the first line, respectively. Rectangle expressing eye position and size, 17e-1
7h is a rectangle representing the positions and sizes of the first to fourth characters from the left of the second line, and 18 shown in FIG. 7C is a table representing the contents of the character array information storage memory 2 by numerical values. , 18a-18h are the position numbers (N
O), the X coordinate (SX) and the Y coordinate (SY) of the upper left corner, the width (W), and the height (H) are expressed by numerical values.
On the other hand, 19 shown in (d) of the figure is a marginal distribution created from the input image 6, 19a and 19b are marginal distributions in the Y direction, 19c and 19b.
d is the marginal distribution in the X direction, and the marginal distribution 19a and X in the Y direction
The peripheral distribution 19c in the direction corresponds to the character pattern on the first line,
The peripheral distribution 19b in the Y direction and the peripheral distribution 19d in the X direction correspond to the character pattern on the second line. Reference numeral 20 shown in FIG. 7E is a rectangle of the character pattern cut out by the priority character cutting means 3a, reference numerals 20a to 20d are rectangles of character patterns corresponding to the rectangles 17c to 17f, and 21 shown in FIG. Is a rectangle of the character pattern cut out by the estimated character cutting-out means 3b.
Is a rectangle of a character pattern cut out corresponding to each of the rectangles 17a, 17b, 17g, and 17h, and 22 shown in FIG. 7G is a rectangle 21 of the character pattern cut out by the estimated character cut-out means 3b.
22a is the peripheral distribution in the Y direction obtained by scanning the input image 16 in the area consisting of a to 21d, where 22a is the rectangle 21a of the character pattern in the first line.
And 21b are the peripheral distributions in the Y direction obtained by scanning the area, and 22b is the peripheral distribution in the Y direction obtained by scanning the area composed of the rectangles 21c and 21d of the second line character pattern, FIG.
23 is a missing hidden area detected based on the value of the peripheral distribution 22, 23a is a hidden area detected based on the value of the peripheral distribution 22a, and 23b is a missing area detected based on the value of the peripheral distribution 22b. is there.

Further, 24 shown in FIG. 9I is a character pattern cut out from the input image 16 in correspondence with the character pattern rectangle 21, and 24a and 24b correspond to the character pattern rectangles 21b and 21d, respectively. In FIG. 25 (j), 25 is the matching pixel with the value "1",
A mask pattern is created for each character pattern 24 with the non-matching pixel value "0" (the black part in the figure is the matching pixel and the white part is the non-matching pixel), and 25a and 25b are the character patterns 24a, respectively. , 24b mask pattern, FIG.
26 and 27 shown in (l) are reference patterns 11a of the character patterns 24, the mask patterns 25 and the categories "0" to "9" of the recognition dictionary 11 shown in FIG.
~ 11j are the candidate categories up to the fifth rank and their similarity obtained by superimposing them with each other, 26 is the candidate category and its similarity to the character pattern 24a and the mask pattern 25a, and 26a to 26e are reference patterns 11e, respectively. , 11a, 11g,
11c, 11i similarity, 27 is a candidate category and its similarity to the character pattern 24b and the mask pattern 25b, 27a
27e are the similarities to the reference patterns 11e, 11g, 11c, 11i, and 11a, respectively. 28 is a recognition category “4” determined for the character pattern 24a from the candidate category and its similarity 26, and 29 shown in FIG.
Is the recognition category "4" determined for the character pattern 24b from the candidate category and its similarity 27. The recognition dictionary stored in the dictionary memory 4 is the same as the recognition dictionary 11 in the operation example of the conventional character reading apparatus shown in FIG.

Next, the operation will be described.

An input image 16 obtained by picking up an image with an image pickup device such as a television camera and performing photoelectric conversion is stored in the image memory 1.

The priority character cutting means 3a scans the input image 16, counts the black pixels in each line, and distributes the margins 19a and 19b in the Y direction.
To create. Then, the marginal distributions 19a and 19b are binarized, and Y
The Y coordinates of both ends of the black portion continuous in the direction are obtained. In the illustrated example, there are two continuous black portions, and two pairs of Y coordinates are extracted. Next, the input image 16 in the range defined by the pair of Y coordinates is scanned for each continuous black portion, and the black pixels of each category are counted to obtain the peripheral distributions 19c and 19d in the X direction. The X-direction marginal distribution 19c is a marginal distribution for the pair of Y coordinates obtained from the Y-direction marginal distribution 19a, and the X-direction marginal distribution 19d is a marginal distribution for the Y-coordinate pair obtained from the Y-direction marginal distribution 19b. Then, the peripheral distributions 19c and 19d in the X direction are set to 2
It is binarized and the X coordinates of both ends of the continuous black portion are obtained. If the broken line in the figure represents a binarization threshold, three pairs of X coordinates are extracted for the marginal distribution 19c, and two pairs of X coordinates are extracted for the marginal distribution 19d. . Next, the pair of Y coordinates for the peripheral distribution 19a in the Y direction and the peripheral distribution 1 in the X direction 1
A pair of X coordinates for 9c is combined to find a rectangle to be attached to the pattern. The rectangles 20a and 20b of the character pattern whose position and size are within a predetermined range are preferentially cut out. The rectangles 20c and 20d of the character pattern are similarly cut out for the peripheral distribution 19b in the Y direction and the peripheral distribution 19d in the X direction.

The estimated character cutout means 3b has a priority character cutout means 3a.
Positions and sizes of the rectangles 20a to 20d of the character pattern cut out in step 17 and the rectangle 17a which is the contents of the character array information storage memory 2
~ 17h and size 18a ~ 18h by comparing the ratio with the size of the rectangles 20a ~ 20d of the character pattern rectangle 17a.
For the remaining rectangles 17a, 17b, 17g, and 17h corresponding to c to 17f, the rectangle 20 of the character pattern that is closest in position respectively
The positions and sizes of the rectangles 21a to 21d of the character pattern are estimated and cut out by multiplying the positions and sizes of c, 20a, 20d, and 20b by the above ratios.

The ratio of the rectangles 21a to 21d to the rectangles 17a to 17h, which are the contents of the character array information storage memory 2, is the priority character cutting means 3a.
Since it is the same as the rectangles 20a to 20d cut out in, it is almost put on the character pattern.

In the missing part detection means 15, the estimated character cutting means 3b
With respect to the rectangles 21a to 21d of the character pattern obtained in step 1, the area consisting of the rectangles 21a and 21b in the same row is scanned, and black pixels are counted for each line to obtain the peripheral distribution 22a in the Y direction. The peripheral distribution 22b in the Y direction is similarly obtained for the rectangles 21c and 21d. Next, the marginal distributions 22a and 22b are binarized with predetermined threshold values, and the Y coordinate of the boundary line between the missing area 23b and the hidden area 23a is obtained. There are two types of threshold values, one for missing detection and the other for hidden detection, but either one is used depending on the shape of the circumscribed rectangle obtained based on the value of the peripheral distribution 19. In the example shown in the figure, the hidden detection is used for the peripheral distribution 22a, and the missing detection is used for the peripheral distribution 22b.

In the character recognition means 5a, with respect to the character patterns of the input image 16 for the rectangles 20 and 21 cut out by the priority character cutout means 3a and the estimated character cutout means 3b, the values of all the pixels that do not overlap with the lacking hidden area 23 are A mask pattern of "1" is created, and other than that, the value of the pixel overlapping with the missing hidden area is set to "0", based on the Y coordinate of the boundary line.
A mask pattern in which the values of the other pixels are "1" is created. For example, for rectangles 21b and 21d of character patterns, mask patterns 25a and 25b are created for the corresponding character patterns 24a and 24b, respectively. Mask pattern 25a
Is a mask of the hidden area. Mask pattern 25
b is a mask of the missing area. Next, the character pattern 24a, the mask pattern 25a, and the reference pattern of each category of the recognition dictionary 11 shown in FIG.
The degree of similarity is obtained by superimposing 11a to 11j on each other. Similarity is similarly obtained for the character pattern 24b and the mask pattern 25b.

In this case, for example, for the input character pattern, a binary matrix in which matching pixels are 1 and non-matching pixels are 0 Is the same as the similarity calculation formula of the conventional character reading device, the similarity is represented by S shown below.

Where the numerator is At the position where the element of is 1 The value obtained by subtracting the number of mismatches from the number of matches of the elements of, the denominator is Is the number of elements with a value of 1. That is, the numerator is a value obtained by subtracting the number of pixels that do not match from the number of pixels that the input character pattern and the reference character pattern match with respect to the pixel whose mask pattern is 1, and the denominator is only the pixels that have a mask pattern value of 1. Is counted.

For example, the similarities 26a to 26e are the reference patterns 11e and 11e, respectively.
a, 11g, 11c, 11i and character pattern 24a and mask pattern 25a
Similarity S calculated by the above equation when and are superposed
Is the value of. Similarly, the similarities 27a to 27e are the similarities when the reference patterns 11e, 11g, 11c, 11i, 11a, the character pattern 24b, and the mask pattern 25b are superimposed.

Next, a category with a high degree of similarity is selected as a candidate category from all the recognition target categories “0” to “9”, and a category satisfying the condition defined by the recognition method from the candidate category and the degree of similarity 26, for example, the degree of similarity The highest category “4” is determined as the recognition category 28. Similarly, the candidate category and similarity 27
For the above, the category “4” is determined as the recognition category 29. Therefore, the recognition result of the character pattern 24a and the recognition result of the character pattern 24b are in category "4".

Incidentally, in the above-mentioned embodiment, the chipped hidden area is obtained only in the Y direction, but it may be obtained in the X direction and the Y direction.

Further, the scanning range for detecting the lacking hidden area has been described as an area formed of the character patterns cut out by the estimated character cutting means 3b, but it may be an area formed of all the cut out character patterns, or the estimated character cutting means 3b. The individual character patterns cut out in step may be used as the scanning range.

Further, as the character recognition method, the case of using the pattern matching method of overlapping patterns has been described, but a method using a feature extracted from the pattern may be used.

〔The invention's effect〕

As described above, according to the present invention, by providing the priority character cutting-out means and the estimated character cutting-out means, it is possible to accurately cut out a character even if the character is missing or hidden due to a change in the imaging direction and position. Can be done. Further, by providing the missing and hidden area detecting means and the character recognizing means for recognizing the missing hidden area, there is an effect that the character can be accurately recognized even when there is a missing or hidden character.

[Brief description of drawings]

FIG. 1 is a block diagram of a character reading device according to an embodiment of the present invention, FIGS. 2 (a) to (n) are diagrams showing an operation example of the device shown in FIG. 1, and FIG. 3 is a conventional character reading device. Device configuration diagram, 4th
(A)-(i) is a figure which shows the operation example of the apparatus shown in FIG. Reference numeral 1 is an image memory, 2 is a character array information storage memory, 3a is a priority character cutting means, 3b is an estimated character cutting means, 5a is a character recognition means, and 15 is a missing area detection means. In the drawings, the same reference numerals indicate the same or corresponding parts.

Claims (1)

[Claims] 1. In a character reading device for reading a character by cutting out and recognizing individual characters whose character arrangement is known from an input image, a character pattern which is obviously a character is prioritized based on a pixel value of the input image. Estimating the character pattern by estimating the position and size of the remaining characters based on the information of the character array and the position and size information of the character pattern cut out by the priority character cutting device. Character cut-out means, a missing hidden area detecting means for detecting an area where a character is missing or hidden based on the pixel value of an area consisting of the character pattern, and an area excluding the missing hidden area detected by the missing hidden area detecting means A character recognition means for recognizing a character by comparing the above-mentioned character pattern with a reference pattern of a character to be recognized. Character reader, characterized in that was e.