JPH1166237A - Optical character reader - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a optical character reader which is shortened in recognition processing time. SOLUTION: A document T is read by an image read part 30 for generating a multi-valued image signal S34. The image signal S34 is stored in an image memory 41, and the image signal S41 outputted from the image memory 41 is converted by a binarizing process part 42 into a binary image signal S42. The binary image signal S42 is filtered by a filter process part 43 to obtain an image signal S43. A character segmenting processing part 44 cuts character pattern areas S44 out of the image signal S43, character by character. For a character pattern area S44, an X-feature quantity extraction part 45a calculates an X-directional feature quantity S45a and a Y-feature quantity extraction part 45c calculates a Y-directional feature quantity S45c. A character discrimination part 46 collates the featured quantities S45a and S45c against the featured quantities of standard patterns so as to discriminate a character, and its code is outputted as a recognition result S46.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an optical character reader (hereinafter referred to as "OCR") having a reduced recognition processing time.

[0002]

2. Description of the Related Art FIG. 2 is a block diagram showing an example of a conventional OCR. In this OCR, the form T is the image reading unit 1
The light is conveyed to a predetermined position by a mechanism (not shown) in FIG. The reflected light from the surface of the form T is imaged on a charge-coupled device sensor (hereinafter, referred to as a CCD sensor) 13 by a lens 12, and is converted into an analog electric signal S13 by the CCD sensor 13. This electric signal S13 is supplied to an analog / digital converter (hereinafter, ADC).
Is converted into a digital image signal S14 of gradation data. The image signal S14 is output to the recognition unit 2
The image data is stored in the image memory 21 in which the image data is 0, and is further binarized by the binarization processing unit 22 into white and black, and the binarization processing unit 22 outputs an image signal S22. Image signal S2
2, the filter processing unit 23 performs a filter process for removing black points (for example, dust) other than the character pattern and white spots in the character pattern. Output signal S of filter processing unit 23
23 is cut out as a character pattern area S24 for each character by a character cutout processing unit 24 and sent to a character feature extraction unit 25.

[0003] A character feature extraction unit 25 extracts a feature of a character conforming to the character recognition method from a character pattern area S24 extracted for each character, and generates an output signal S25. The character identification unit 26 performs a character recognition process of determining a character by comparing a dictionary in which a standard pattern is stored in advance with the output signal S25, and outputs a corresponding character code as a recognition result S26. These image memories 2
1, binarization processing unit 22, filter processing unit 23, character extraction processing unit 24, character feature extraction unit 25, and character identification unit 26
Is controlled by the recognition control unit 27. FIG. 3 shows a multi-valued (for example, 8 gradation) stored in the image memory 21 in FIG.
It is a figure which shows the example of the character pattern of. In this figure,
Represents white, and “0” represents black. “1” to “6” are between white and black, and the larger the numerical value, the closer to white.

FIG. 4 shows the multi-valued character pattern shown in FIG.
2 shows an example of a character pattern that has been subjected to a binarization process, a filter process, and a process of extracting each character by a binarization processing unit 22, a filter processing unit 23, and a character extraction processing unit 24 in FIG. FIG. This figure shows an example in which the binarization threshold is set to, for example, “3”. Therefore, “·” to “4” in FIG. 3 are white, and “0” to “3” are black. The character pattern extracting unit 25 and the character identifying unit 26 perform a character recognition process on the character pattern area S24 extracted in units of one character as described above. In this character recognition processing, as a character recognition method for numbers, alphabets, and katakana, a stroke analysis method for analyzing a character stroke, a pattern matching method for checking a match between a character pattern and a reference pattern, and Various methods have been proposed as a method of absorbing character deformation, such as a method of analyzing a line structure of a character pattern and a method of analyzing a background structure thereof.

[0005]

However, the OCR of FIG. 2 has the following problems. When using the stroke analysis method as the character recognition method, for example,
Since the characteristics of each character are described in the dictionary including the deformation,
This dictionary will be huge. For this reason, there is a problem that the processing time required for the recognition becomes longer, and if the processing time is reduced, the scale of the hardware becomes larger. In the method of analyzing the background structure, a black-and-white point forming the character is determined by focusing on the background portion of the character, and a loop of a line forming the character, a concave state, a convex state, and the like are extracted as features. However, the character is recognized based on this feature, but there is a problem that the process of extracting this feature is complicated.

[0006]

According to a first aspect of the present invention, a multi-valued image signal is read by reading an image of a character on a form using an optical system in an OCR. An image reading unit for generating, a binarization processing unit for binarizing the multi-level image signal to generate a binary image signal, and a character extracting unit for extracting a character pattern area for each character from the binary image signal A processing unit, an X feature amount calculation unit that scans the character pattern area in the X direction of XY coordinates and calculates the number of continuous black points in the X direction of the character pattern in the character pattern area as an X direction feature amount; A Y feature amount calculation unit that scans the character pattern area in the Y direction of the XY coordinates and calculates the number of continuous black points in the Y direction of the character pattern in the character pattern area as a Y direction feature amount; Features and Y-direction features Has a prestored dictionary the reference pattern of the X-direction feature amount and the Y-direction feature amount and a character identifying unit that identifies the characters on the form by comparing it with the reference pattern comprises. By employing such a configuration, an image of a character on a form is read by the image reading unit, and a multi-valued image signal is generated. This multi-level image signal is binarized by a binarization processing unit and output as a binary image signal. This binary image signal is
In the character extraction processing unit, each character is extracted into a character pattern area. The character pattern area is scanned in the X direction and the Y direction by the X feature value calculation unit and the Y feature value calculation unit, respectively, and the X direction feature value and the Y direction feature value are calculated. The X-direction feature amount and the Y-direction feature amount are identified by comparison with a standard pattern stored in a dictionary of the character identification unit.

According to a second aspect of the present invention, in the OCR, the image reading unit, the binarization processing unit, and the character cutout processing unit according to the first invention, and the character pattern area are normalized to a predetermined size. A normalization processing unit that generates a character pattern area; and scans the normalized character pattern area in the X direction of XY coordinates to determine the number of continuous black points of the character pattern in the normalized character pattern area in the X direction. An X feature amount calculation unit for calculating as a feature amount, and the normalized character pattern area is defined as Y of the XY coordinates.
A Y feature amount calculating unit that scans in the direction to calculate the number of continuous black points in the Y direction of the character pattern in the normalized character pattern area as a Y direction feature amount; A dictionary that stores a standard pattern in advance, and a character identification unit that identifies a character on the form by comparing the X-direction feature amount and the Y-direction feature amount with the standard pattern.
Have. By adopting such a configuration, the character pattern area extracted by the character extraction processing unit of the first invention is normalized to a predetermined size by the normalization processing unit and becomes a normalized character pattern area. Is output. The normalized character pattern area is scanned in the X direction and the Y direction by the X feature amount calculation unit and the Y feature amount calculation unit, respectively, and the X direction feature amount and the Y direction feature amount are calculated. The X-direction feature amount and the Y-direction feature amount are identified by comparison with a standard pattern stored in a dictionary of the character identification unit. Therefore, the above problem can be solved.

[0008]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS First Embodiment FIG. 1 is a block diagram of an OCR showing a first embodiment of the present invention. This OCR has an image reading unit 30. The image reading unit 30 reads an image of a character on the form T illuminated by the light source 31 using an optical system (for example, a lens 32 and a CCD sensor 33), and outputs an analog output signal S33 of the CCD sensor 33 to the ADC 34. It has a function of generating a digital multi-level image signal S34. A recognition unit 40 is connected to the output side of the ADC 34. The recognizing unit 40 has an image memory 41 for storing the multi-valued image signal S34. The output side of the image memory 41 converts the multi-valued image signal S41 stored in the image memory 41 into a binary image. Signal S
The binarization processing unit 42 for generating the data 42 is connected. 2
On the output side of the binarization processing unit 42, a binary image signal S42
To the binary image signal S4
3 is connected. On the output side of the filter processing unit 43, the binary image signal S4
3 is connected to a character extraction processing unit 44 for extracting a character pattern area S44 for each character. On the output side of the character extraction processing unit 44, the character pattern area S44 is scanned in the X direction of the XY coordinates, and the X of the character pattern in the character pattern area S44 is scanned.
The number of continuous black points in the direction is counted, and the X-direction feature amount S4 is calculated.
An X feature amount calculation unit 45a for calculating as 5a is connected. The X-direction feature value S45a is stored in the X memory 45b. On the output side of the character extraction processing unit 44, the character pattern area S44 is scanned in the Y direction of the XY coordinates, and the number of continuous black points of the character pattern in the character pattern area S44 in the Y direction is counted. Direction feature value S45
The Y feature amount calculation unit 45c that calculates as c is connected. The Y direction feature amount S45c is stored in the Y memory 45d. A character identification unit 46 is connected to each output side of the X feature amount calculation unit 45a and the Y feature amount calculation unit 45c. The character identification unit 46 determines the X-direction feature amount S45.
a dictionary that previously stores standard patterns of “a” and “Y” direction feature values S45c, and has a function of identifying a character on the form T by comparing the “X” direction feature value S45a and the “Y” direction feature value S45c with the standard pattern. Have. Each block in the recognition unit 40 is controlled by a recognition control unit 47.

Next, the operations (1) to (3) of the OCR shown in FIG.
Will be described. (1) Overall operation of the OCR in FIG. 1 In this OCR, a form T is conveyed by a mechanism (not shown) in the image reading unit 30 and a light source 31 at a predetermined position.
Is irradiated. The reflected light from the surface of the form T is imaged on the CCD sensor 33 by the lens 32, and the CCD sensor 33
Is converted into an analog electric signal S33. The electric signal S33 is converted by the ADC 34 into a multi-valued image signal S34.
Is converted to The image signal S34 is stored in an image memory 41 in the recognizing unit 40, and the image signal S41 output from the image memory 41 is converted into a binary image signal S42 based on a threshold set in advance by a binarization processing unit 42.
Is converted to The binary image signal S42 is subjected to a filtering process for removing noise such as dust in a filtering section 43, and the filtering section 43 outputs an image signal S43. The character cutout processing section 44 outputs the image signal S43.
, A character pattern area S44 is cut out for each character. For the character pattern area S44, the X feature amount calculation unit 45a
The feature amount S45a in the direction is calculated, and then the Y feature amount calculating unit 45c calculates the feature amount S45c in the Y direction. The character identifying unit 46 identifies the character by comparing the characteristic amounts S45a and S45c with the characteristic amount of the dictionary in which the standard pattern is stored in advance, and outputs the character code as the recognition result S46.

(2) Calculation of X and Y feature values FIG. 5 is a diagram showing the concept of calculating feature values in the OCR shown in FIG. 1, and FIG. 6 is a diagram for explaining an example of calculating feature values. As shown in FIG. 5, the X feature quantity calculation unit 45a scans the first row in the Y direction in the X direction, counts the number of continuous black points in the X direction, and stores the number of continuous black points in the X memory 45b. Is stored. In FIG. 5, (1, 1) is stored in the X memory 45b because the number of consecutive black dots is made up of two blocks of (1, 1). Next, the X feature amount calculation unit 45a
Scans the second row in the Y direction in the X direction, counts the number of continuous black points in the X direction, and stores (3) in the X memory 45b. Similarly, the X feature quantity calculation unit 45a counts the number of continuous black points in the X direction up to the fifth row in the Y direction, and (1), (3, 1), and (2, 2) indicate the X direction. The feature amount S45a is sequentially stored in the X memory 45b. On the other hand, the Y feature amount calculation unit 45c scans the first column in the X direction in the Y direction, counts the number of continuous black points in the Y direction, and stores the number of continuous black points in the Y memory 45d.
In the first column, (1, 2) is stored in the Y memory 45d because the number of consecutive black points is composed of two blocks (1, 2). Similarly, the Y feature amount calculation unit 45c
Counts the number of continuous black points in the Y direction up to the fifth column in the X direction, and (2), (3), (2, 1), and (1, 2) indicate Y direction feature amount S45d as Y Memory 45
d.

FIG. 6 shows the number of continuous black spots in the X and Y directions in the pattern of FIG. Characters are determined by comparing the number of these black points with the X-direction feature amount S45a and the Y-direction feature amount S45d calculated by the method shown in FIG. 5 against the dictionary feature amount stored in the standard pattern. You. Note that the standard pattern for each category is also X
The direction and the Y direction are independent, and similarly to the X memory 45b and the Y memory 45d, the X direction feature amount is sequentially described from the first row in the Y direction, and the Y direction feature amount is described from the first column in the X direction. They are described sequentially. This collation method uses the X memory 45b.
And the heads of the Y memory 45d and the heads of the standard patterns in the X and Y directions of each category are sequentially collated, and the number of mismatches between the number of black points in the X direction and the number of black points in the Y direction are counted. The character having the smallest total number of mismatches in the X direction and the Y direction is regarded as the recognition result S46. Here, the X memory 45b and the Y memory 45d to be compared may not have the same number of blocks as the standard pattern. For this reason, the positions of the blocks with the largest number of consecutive black dots are matched, and the number of mismatches of the number of black dots is counted. Even when the number of blocks matches, the position of the block with the largest number of consecutive black points is matched, and the number of mismatches of the number of black points is counted.

(3) Character identification operation FIG. 7 is a diagram showing an example of a method of comparing a memory with a standard pattern in the character identification section 46 in FIG. This FIG.
As shown in (1), the memory M, which is the X memory 45b or the Y memory 45d, and the standard pattern SP are sequentially collated from the head. That is, three blocks (1, 5, 3) in the memory M are compared with two blocks (6, 3) in the standard pattern SP. In this case, if the position of “5” in the memory M, which is the block with the largest number of continuous black dots, matches the position of “6” in the standard pattern, and the number of mismatches in each block is counted,
The number of mismatches becomes "1", "1", and "0", respectively, and the total number of mismatches becomes "2". Next, when the memory M and the standard pattern SP have the same number of blocks as 2, as in (1,3) and (3,1), respectively, the position of each "3" in the memory M and the standard pattern SP Are matched, and the number of mismatches in each block is counted. The numbers of mismatches are "1", "0", and "1", respectively, and the total number of mismatches is "2". Similarly, when the memory M and the standard pattern SP are (1, 3, 1) and (3, 1, 2), respectively,
When the number of blocks matches 3, the position of each "3" of the memory M and the standard pattern SP is matched, and the number of mismatches of each block is counted. The numbers of mismatches are "1" and "0", respectively. , "0", "2", and the total number of mismatches becomes "3".

Further, when the memory M and the standard pattern SP are two blocks as shown in (3, 2) and (4, 2), the position of "3" of the memory and the position of "4" of the standard pattern are determined. When the number of inconsistencies in each block is counted in advance, the numbers of inconsistencies become "1" and "0", respectively, and the total number of inconsistencies becomes "1". When the memory M and the standard pattern SP are (3) and (3, 3), respectively, the number of mismatches is “3”. Thus, the memory M and the standard pattern SP
Is completed, the number of mismatches as characters is "2", "2", "3", "1", "3", respectively, and the total number of mismatches is "11". . In this manner, in the X direction and the Y direction, the memory M is compared with all the categories of the standard pattern SP, and a character having the smallest total number of mismatches in the X direction and the Y direction is determined as an identification result.
Here, when comparing the memory M with the standard pattern SP, the size of the character becomes a problem. In other words, if the vertical and horizontal sizes of the characters are different, the collation will be performed not on the row or column that should be collated but with the shifted row or column. Make sure you have a dictionary that is large,
What is necessary is just to match with the large standard pattern.

As described above, in the first embodiment,
There are the following advantages. (A) X feature value calculation unit 45a and Y feature value calculation unit 45
c is the X-direction feature amount S45a and the Y-direction feature amount S4 of the character.
5d is calculated and stored in the X memory 45b and the Y memory 45d, respectively, and is compared with the standard pattern SP by the character identification unit 46 to extract the characteristic of the character. And the feature extraction process can be made very simple. (B) When the characters to be identified are type numbers, alphabets, or katakana, since the size of the same type of character has little variation, the sizes of various characters stored in the dictionary as the standard pattern SP are X direction and Y direction. About two or three types of directions are sufficient. Because the dictionary does not become huge,
Recognition processing time can be reduced as compared with the conventional case.

Second Embodiment FIG. 8 is a block diagram of an OCR according to a second embodiment of the present invention. In FIG. 8, the same elements as those in FIG. Is attached. In this OCR, a recognition unit 40A having a different configuration is used instead of the recognition unit 40 in FIG.
Is provided. In the recognizing unit 40A, the output side of the character cutout processing unit 44 in FIG.
A new normalization processing unit 48 is provided between the input side of the feature amount extraction unit 45c and the recognition control unit 47 in FIG.
A recognition control unit 47A having a configuration different from that of the first embodiment is provided. The normalization processing unit 48 has a function of normalizing the character pattern area S44 extracted from the character extraction processing unit 44 to a predetermined size to generate a normalized character pattern area S48.
Each block in the recognition unit 40A is controlled by a recognition control unit 47A. Other configurations are the same as those in FIG.

Next, the operations (1) and (2) of the OCR shown in FIG.
Will be described. (1) Overall operation of the OCR in FIG. 8 The operation of the OCR differs from FIG. 1 in the following points.
The character pattern area S44 for each character extracted by the character extraction processing section 44 is normalized by the normalization processing section 48 in the vertical and horizontal directions to a predetermined size, and the normalization processing section 48 outputs the normalized character pattern area S48. Is output. With respect to the normalized character pattern area S48, the X feature amount extraction unit 45a outputs the X direction feature amount S
45a, and then the Y feature amount extraction unit 45c calculates the Y direction feature amount S45c. The character identification unit 46 identifies the character by comparing the characteristic amounts S45a and S45c with the characteristic amount of the dictionary stored in advance in the standard pattern, and outputs the character code as the recognition result S46.

(2) Normalization Processing Operation FIG. 9 is a diagram showing an example of a pattern obtained by normalizing FIG. In FIG. 9, 22 pixels in the vertical direction and 1 pixel in the horizontal direction in FIG.
The pattern of three pixels is normalized to eight pixels in the vertical direction and eight pixels in the horizontal direction. The X feature quantity extraction unit 45a and the Y feature quantity extraction unit 45c perform the same operation as in the first embodiment. That is, the X feature quantity extraction unit 45a first counts the number of black points continuous in the X direction from the first row to the eighth row in the Y direction of the pattern shown in FIG. 9, and stores the number in the X memory 45b. next,
The Y feature quantity extraction unit 45c counts the number of black points continuous in the Y direction from the first column to the eighth column in the X direction, and
5d. In the case of the character “2” in FIG. 9, the X memory 45b stores (7), (2),..., (1, 3), (5)
Are sequentially stored, and (3, 1) are stored in the Y memory 45d.
(3, 1), (1, 2, 1), ..., (1, 3), (2)
Are sequentially stored. Similarly, in the case of the character “3”, (7), (2),..., (3) and (7) are sequentially stored in the X memory 45b, and (1, 1), (7) are stored in the Y memory 45d. 1,
.., (4, 2), (3) are sequentially stored.

The collation between the memory and the standard pattern is performed in the same manner as in the first embodiment. That is, X memory 45b and Y memory
The memory 45d is sequentially compared with the standard pattern in the X direction and the Y direction of each category from the beginning, and the number of mismatches in the number of black points in the X direction and the Y direction is counted. Characters whose total number is the smallest are identified as result S46
And In the present embodiment, when the memory and the standard pattern are collated, the character size is normalized, so that the character identification unit 46 need only collate with one standard pattern in each of the X direction and the Y direction. . As described above, in the second embodiment, the normalization processing unit 48 determines the character pattern area S44 for each character extracted by the character extraction processing unit 44.
Is normalized, the size of various characters stored in the dictionary as a standard pattern may be one type for each of the X direction and the Y direction. Therefore, the dictionary can be made very small, and in addition to the advantages of the first embodiment, the recognition processing time can be further reduced.

The present invention is not limited to the above embodiment,
Various modifications are possible. For example, there are the following modifications. (A) In the second embodiment, when the size of the character pattern area S44 is normalized, the size is normalized to 8 pixels in the vertical direction and 8 pixels in the horizontal direction. The ratio of the number of pixels need not be the same, and may be normalized to any size as long as a feature amount necessary for character recognition can be obtained and each category can be identified. (B) The values of the memory M and the standard pattern SP shown in FIG.
Other values may be used.

[0020]

As described above in detail, according to the first aspect, only the number of continuous black spots in the X and Y directions of the character pattern extracted by the character extraction processing section is counted. Since the feature is extracted, there is no need to perform complicated feature extraction as in the related art, and the feature extraction processing can be very easily performed. Furthermore, when the characters are type numbers, alphabets, or katakana, the sizes of the same type of characters are less scattered. Therefore, the sizes of various characters stored in the dictionary as standard patterns are 2 to 2 in the X and Y directions, respectively. About three types are sufficient. For this reason, the dictionary does not become enormous, so that the recognition processing time can be reduced as compared with the related art. According to the second aspect, the normalization processing unit normalizes the character pattern area for each character extracted by the character extraction processing unit, so that the size of various characters stored in the dictionary as a standard pattern is determined. One type may be used for each of the X direction and the Y direction. Therefore, the dictionary can be made very small, and in addition to the effects of the first invention,
Further, the recognition processing time can be reduced.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of an OCR according to a first embodiment of the present invention.

FIG. 2 is a configuration diagram of a conventional OCR.

FIG. 3 is a diagram showing an example of a multi-valued character pattern in an image memory 21 in FIG. 2;

FIG. 4 is a diagram illustrating an example of a character pattern extracted by binarizing the multi-valued character pattern in FIG. 3;

FIG. 5 is a diagram showing a concept of feature amount calculation in the OCR of FIG. 1;

FIG. 6 is a diagram illustrating calculation of a feature amount.

FIG. 7 is a diagram showing a comparison between a memory and a standard pattern in the OCR of FIG. 1;

FIG. 8 is a configuration diagram of an OCR according to a second embodiment of the present invention.

FIG. 9 is a diagram showing a pattern obtained by normalizing FIG. 4;

[Explanation of symbols]

 Reference Signs List 30 image reading section 31 light source 32 form T lens 33 CCD sensor 34 A / D converter 40, 40A recognition section 41 memory 42 binarization processing section 43 filter processing section 44 character extraction processing section 45a X feature quantity extraction section 45c Y Feature amount extraction unit 46 Character identification unit 47 Recognition control unit 48 Normalization processing unit

Claims (2)

[Claims] An image reading unit that reads a character image on a form using an optical system to generate a multi-valued image signal; and a binary that generates a binary image signal by binarizing the multi-valued image signal. A character pattern processing unit that cuts out a character pattern area for each character from the binary image signal; and scans the character pattern area in the X direction of the XY coordinates to extract a character pattern in the character pattern area. the Y of the X feature amount calculation unit for calculating a continuous black points of the X-direction as the X direction feature quantity, by scanning the character pattern area in the Y direction of the XY coordinates of the character pattern of the character pattern area A Y feature amount calculation unit for calculating the number of black spots in the direction as a Y direction feature amount; a dictionary in which standard patterns of the X direction feature amount and the Y direction feature amount are stored in advance; Directional features are calculated using the standard pattern Optical character reader, characterized in that the character identifying unit that identifies the characters on the form, with the by comparing the. 2. The image reading unit according to claim 1, a binarization processing unit, and a character extraction processing unit, and a normalization unit that normalizes the character pattern area to a predetermined size to generate a normalized character pattern area. An X feature that scans the normalized character pattern area in the X direction of XY coordinates and calculates the number of continuous black points in the X direction of the character pattern in the normalized character pattern area as an X direction feature amount. An amount calculating unit that scans the normalized character pattern area in the Y direction of the XY coordinates and calculates the number of continuous black points in the Y direction of the character pattern in the normalized character pattern area as a Y direction feature amount. A feature value calculation unit; and a dictionary in which standard patterns of the X-direction feature value and the Y-direction feature value are stored in advance, and comparing the X-direction feature value and the Y-direction feature value with the standard pattern, Character identification to identify the character of Preparative, optical character reading apparatus characterized by comprising.