JPH11161739A - Character recognizing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a character recognizing device for preventing misreading for a void character. SOLUTION: A picture storing part 1 stores a binary picture image ST of a medium T by using a background as white picture elements and a character as black picture elements. When the binary picture image S is stored in the picture storing part 1, a character area is segmented by one character unit from the binary picture image S1 by a character segmenting part 2, and the segmented character area S2 is transmitted to a character discriminating part 3 and a character recognizing part 4. The character judging part 3 judges whether or not the character area S2 is a normal character or a void character. The character judging part 3 judges a character, and transmits a character judged result S3 to the character recognizing part 4. The character recognizing part 4 receives the judged result S3, receives the character area S2 from the character segmenting part 2, and when the judged result S3 indicates the normal character, the character recognizing part 4 operates a recognition processing, and outputs a recognized result S4. Also, when the judged result S3 indicates the void character, the character recognizing part 4 outputs the recognized result S4 as reject (unreadable) without operating any recognition processing.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition apparatus (hereinafter referred to as "OCR") which cuts out and recognizes a character from a binary image of a document or the like, and particularly takes measures against the recognition of an outline character.
It is about).

[0002]

2. Description of the Related Art A conventional OCR is, for example, a CCD (Charge).
A document or the like is optically read by using a sensor or the like, an image image quantized into binary values of white and black is obtained, and a character area is cut out from the image image to perform recognition processing. In this recognition processing, a recognition dictionary in which features are extracted from a predetermined standard character pattern (hereinafter, referred to as a standard pattern) is compared with the features of the character pattern to be recognized. Select and output as recognition result.

[0003]

However, the conventional OCR has the following problems. FIG. 2 is a diagram illustrating an example of a normal character. In the recognition dictionary, the standard pattern is created using a character composed of only black pixels as shown in FIG. 2 (this is called a normal character). This is because ordinary characters are frequently used in general documents. However, with today's word processors and the like, multi-functions and increased disk capacities make it possible to easily create documents using blank characters, various fonts, decorative characters, and the like. FIG. 3 is a diagram illustrating an example of outline characters. As shown in FIG. 3, a document may be created using characters in which only the edges of the characters are composed of black pixels (ie, white characters). In addition, when there is a word or the like to be particularly emphasized in a document, these characters may be used to distinguish them from other characters. Conventional O
In CR, when these white characters are recognized using a recognition dictionary for ordinary characters, the characteristics are significantly different from those of ordinary characters, so that the probability of misreading is high and the recognition accuracy is reduced. There was a problem. Further, if features of white characters are added to the recognition dictionary of normal characters in order to make white characters the target of recognition, there is a problem that the recognition accuracy for the normal characters is reduced. For this reason, there is a demand for an OCR that does not reduce the recognition accuracy even when white characters are included in a document or the like to be recognized.

[0004]

According to a first aspect of the present invention, there is provided an image storage unit for inputting and storing a binary image of a medium on which characters are written, In an OCR having a character extracting unit for extracting a character area from the binary image stored in the image storage unit and a character recognizing unit for recognizing a character in the character area, the following means is provided. Is provided. The character in the cut-out character area is a normal character in which a character line forming the character is composed only of black pixels or an outline character in which only the edge of the character line is composed of black pixels. Is provided. Further, the character recognition unit is configured not to recognize the character when the result of the determination by the character determination unit indicates the outlined character. According to a second aspect, in the OCR, the same image storage unit as the first aspect,
A character extraction unit, a character discrimination unit, a normal character dictionary unit storing a recognition dictionary suitable for recognition of the normal characters, and a white character dictionary unit storing a recognition dictionary suitable for recognition of the white characters A dictionary selecting unit that selects the normal character dictionary unit or the outline character dictionary unit based on a result of the determination by the character determining unit; and a character recognition unit that recognizes characters from the character region using a recognition dictionary selected by the dictionary selection unit. And a character recognition unit that performs

In a third aspect of the present invention, the character discriminating unit of the first or second aspect of the present invention includes the normal contour counting pixels constituting the contour outside the black pixel area of the character in the cut-out character area. The number of points and the number of loop contour points obtained by counting the pixels constituting the contour inside the black pixel area of the character are detected, and the ratio between the normal contour points and the loop contour points is compared with a predetermined threshold value. It is configured to determine whether the character is the ordinary character or the outline character based on the comparison result. According to the first and third aspects, since the OCR is configured as described above, the binary image of the medium is stored in the image storage unit,
A character extracting section extracts a character area from the stored binary image. The characters in the extracted character area are
The character determining unit determines whether the character is a normal character or a white character. The character recognition unit recognizes the character when the result of the determination by the character determination unit indicates a normal character, and does not recognize the character when the result of the determination indicates a blank character. Second and third
According to the invention, the dictionary selection unit selects the normal character dictionary unit or the outline character dictionary unit based on the determination result of the character determination unit. The character recognizing unit recognizes a character in the character area cut out from the character cutout unit based on the recognition dictionary of the selected dictionary unit.

According to a fourth aspect of the present invention, in the OCR, an image storage unit for inputting and storing a binary image image of a medium on which characters are written, and a character area from the binary image image stored in the image storage unit. Character extraction portion, and the characters in the extracted character area, the character line forming the character is a normal character composed of only black pixels or only the edge of the character line is a black pixel A character discriminating unit for discriminating whether the character is a white character, and when the discrimination result of the character discriminating unit indicates the normal character, the character in the cut-out character area is output as it is, and the discrimination result When the character indicates the white characters, an image correction unit that corrects and outputs the characters in the cut-out character area to the normal characters, and a character recognition unit that recognizes the characters output from the image correction unit , Have. In a fifth aspect, the character discriminating unit according to the fourth aspect comprises a normal contour point number obtained by counting pixels constituting a contour outside a black pixel area of the character in the cut-out character area, and coordinates of the pixel. Is detected, and the number of loop contour points obtained by counting the pixels constituting the contour inside the black pixel area of the character is detected, and the ratio between the normal contour points and the loop contour points is determined by a predetermined threshold value. A comparison is made to determine whether the character is the normal character or the outline character based on the comparison result, and further, the normal contour data is output. Sixth
In the OCR according to the present invention, in the OCR, an image storage unit for inputting and storing a binary image image of a medium on which characters are written, and a character cutout for extracting a character area from the binary image image stored in the image storage unit When the determination result of the output unit, the character determination unit of the fifth invention, and the character determination unit indicates the normal character, the character in the cut-out character area is output as it is, and the determination result is the white character. In the case of indicating a blank character, the image processing device includes an image correction unit that corrects and outputs the character in the cut-out character area to the normal character, and a character recognition unit that recognizes a character output from the image correction unit. .

The image correction section receives the character area from the character cutout section, receives the determination result and the normal outline data from the character determination section of the fifth invention, and determines the normal character as the result of the determination. A switching unit that sends the character area to the character recognition unit when the character area is indicated, and sends the normal outline data and the character area to the outline rectangle detection unit when the determination result indicates the outline character; The minimum rectangle surrounding the contour is calculated from the normal contour data, and the orthogonal X axis and Y
As a result of comparing the calculated coordinate values specifying the position and size of each of the calculated rectangles in the XY coordinate system composed of axes, when a certain rectangle completely surrounds another rectangle, the enclosed rectangle is formed. The contour rectangle detecting section which ranks the respective contour data by positioning the normal contour data at the top and the normal contour data forming the enclosed rectangle at the bottom, Received each of the normal contour data and the character area, and for the pixels of each area surrounded by each of the normal contour data in the character area, if the rank is an odd rank, convert it to a black pixel; When the rank is an even rank, a process of converting to white pixels is performed in order from the top, thereby correcting a character in the character area to the normal character and transmitting the corrected image to the character recognition unit. , Have According to the fourth, fifth, and sixth aspects of the present invention, when the determination result of the character determination unit indicates a normal character, the image correction unit outputs the character in the character region cut out by the character cutout unit as it is. . When the result of the determination by the character determination unit indicates an outline character, the image correction unit converts the character in the character area into a normal character so that the character recognition unit can recognize the character using a recognition dictionary suitable for recognition of the normal character. Correct and output. The character output from the image correction unit is recognized by the character recognition 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. The OCR has an image storage unit 1 for inputting and storing a binary image ST of a medium T on which characters are described. On the output side of the image storage unit 1, the image storage unit 1
Is connected to a character extracting section 2 for extracting a character area from the binary image image S1 stored in the storage area. On the output side of the character extracting unit 2, a character determining unit 3 for determining whether the character in the extracted character area is a normal character or an outline character is connected. Further, on the output side of the character extracting unit 2, when the determination result in the character determination unit 3 is a normal character, the character in the character area is recognized and when the determination result is a blank character, a reject ( A character recognizing unit 4 for outputting as (unread) is connected. The character recognition unit 4 outputs a character recognition processing result or rejection as a recognition result S4. Next, processing in the OCR of FIG. 1 will be described. The image storage unit 1 stores the binary image ST of the medium T, for example, as white pixels for the background and black pixels for the characters. When the binary image ST is stored in the image storage unit 1, the character extracting unit 2 extracts a character area from the binary image S1 in units of one character, and extracts the extracted character area S2 into the character discriminating unit 3 and It is sent to the character recognition unit 4. The character determination unit 3 determines whether the character area S2 is a normal character or a white character.

Hereinafter, the determination methods (1) and (2) will be described. (1) Detection of Character Outline First, the character area S2 is scanned to detect the character outline.
To detect the contour, a contour tracking method described below is used. FIGS. 4A and 4B are diagrams illustrating the contour tracking method. In the contour tracing method, as shown in FIG. 4A, the upper left of the character area S2 is set as a scanning start point s, and the character area S2
Are sequentially scanned. At this time, when a black pixel in contact with a white pixel is detected, as shown in FIG. 4B, the contour is detected by following the black pixel in a counterclockwise direction. A black pixel in contact with a white pixel refers to a pixel in which a white pixel exists at any of the upper, lower, left, and right sides of the black pixel. Specifically, a black pixel that is in contact with a white pixel detected first in the scanning process is set as a contour tracking start point, and this is set as a black pixel b0. Next, the black pixel b1 is detected in the counterclockwise direction starting from the black pixel b0. When the black pixel b1 is detected, the black pixel b2 is detected counterclockwise around the black pixel b1 and starting at the black pixel b0. As described above, the detection of the black pixel is repeated until the contour is traced and the pixel returns to the black pixel b0 at the start point. Then, the number of tracked black pixels is written in the table in the character discriminating unit 3 as the number of contour points. At this time, the same contour is not tracked again.

(2) Character determination FIGS. 5A and 5B show outline data of a normal character.
FIGS. 6A and 6B are diagrams showing outline data of outline characters. The contour data includes normal contour data configured to surround the area of the black pixel and loop contour data configured inside the area of the black pixel. In order to distinguish them, the case where the pixel immediately above the start point of the contour tracking is a white pixel is defined as normal outline data, and the case where the pixel above the start point of the outline tracking is a black pixel is defined as loop outline data. I do. When the pixel immediately above the start point of the contour tracking is outside the character area, it is determined to be normal contour data. In the character discriminating unit 3, when tracking the contour,
Usually, the number of contour points and the number of loop contour points are counted. In general,
A normal character has a feature that the number of loop contour points is smaller than that of a normal contour point, and a blank character has a feature that the number of normal contour points and the number of loop contour points do not change much. Therefore, the number of normal contour points is A, the number of loop contour points is B, and the contour data ratio C is obtained by the following equation (1). C = A / B (1) Since the contour data ratio C is large for normal characters and small for outline characters, the character discriminating unit 3 sets the contour data ratio C to a predetermined threshold value THL. If the following expression (2) is satisfied, the character is determined as a normal character, and if not, the character is determined as a blank character. C> THL (2) In the outline data of the normal character shown in FIGS. 5A and 5B,
The number of normal contour points A = 275 in FIG.
Is obtained, and the contour data ratio C = 3.35 is obtained. In the outline data of the outline characters shown in FIGS. 6A and 6B, the number of normal outline points A = 374 in FIG. 6A and the number of loop outline points B = 34 in FIG.
2 is obtained, and the contour data ratio C = 1.09 is obtained. Therefore, if the threshold value THL is set to, for example, 2 in advance, it is possible to distinguish between a normal character and a white character. still,
When the loop contour number B is 0, the character is determined as a normal character without using the expression (1).

When the character discrimination section 3 discriminates a character, it sends a character discrimination result S3 to the character recognition section 4. The character recognition unit 4 receives the determination result S3, receives the character area S2 from the character cutout unit 2, and performs a recognition process when the determination result S3 indicates a normal character, and outputs the recognition result S4. or,
When the discrimination result S3 indicates an outline character, the character recognizing unit 4 outputs the recognition result S4 as reject (unread) without performing the recognition process. As described above, in the first embodiment, even when a document including white characters is to be recognized, the character determination unit 3 determines whether the character is a normal character or a white character, which may result in erroneous reading. Since the white characters having high recognizability are not read, the recognition accuracy does not decrease. Further, since the recognition result S4 is output as a reject, the operator can easily confirm the recognition result, perform a correction operation, and the like in the processing by the correction terminal in the subsequent stage.

Second Embodiment FIG. 7 is a block diagram of an OCR according to a second embodiment of the present invention, in which elements common to those in FIG. 1 showing the first embodiment have common reference numerals. Is attached. In this OCR, a character recognition unit 4A having a different configuration is connected to the output side of the character discrimination unit 3 instead of the character recognition unit 4 in FIG. Further, a dictionary selection unit 5 is connected to an output side of the character discrimination unit 3, and the dictionary selection unit 5 includes a normal character dictionary unit 6 for storing a normal character dictionary created by extracting features from normal characters.
An outline character dictionary unit 7 that stores an outline character dictionary created by extracting features from outline characters is connected.
When the determination result S3 of the character determination unit 3 is a normal character, the dictionary selection unit 5 selects the normal character dictionary unit 6, and determines the determination result S3.
If the character 3 is an outline character, the outline character dictionary unit 7 is selected. A character recognition unit 4A is connected to an output side of the dictionary selection unit 5. The character recognizing unit 4A is provided with the normal character dictionary unit 6 or the white character dictionary unit 7 selected by the dictionary selecting unit 5.
Is for recognizing characters from the character area S2 based on the recognition dictionary. Other configurations are the same as those in FIG.

Next, the processing in the OCR of FIG. 7 will be described. Image storage unit 1, character extraction unit 2, and character determination unit 3
Performs the same processing as in the first embodiment. Dictionary selection unit 5
Receives the discrimination result S3 of the character area S2 from the character discrimination unit 3, connects to the normal character dictionary unit 6 when the discrimination result S3 indicates a normal character, and reads the recognition dictionary stored in the normal character dictionary unit 6. It is sent to the character recognition unit 4A. Also, the determination result S3
Indicates white characters, the dictionary selection unit 5 connects to the white character dictionary unit 7 and sends the recognition dictionary stored in the white character dictionary unit 7 to the character recognition unit 4A. Character recognition unit 4A
Receives the character area S2 from the character extracting section 2, receives the dictionary from the dictionary selecting section 5, performs recognition processing of characters in the character area S2, and outputs a recognition result S4A. As described above, in the second embodiment, the normal character dictionary unit 6 and the outline character dictionary unit 7 are prepared in advance, and even when a document in which outline characters exist is to be recognized, the character discrimination is performed. The unit 3 determines whether a character is a normal character or an outline character, and the dictionary selecting unit 5 selects an optimal dictionary according to the determination result S3 and sends it to the character recognizing unit 4A. Is performed as before, and white characters can be correctly recognized.

Third Embodiment FIG. 8 is a block diagram of an OCR according to a third embodiment of the present invention. Elements common to the elements in FIG. 1 according to the first embodiment are denoted by the same reference numerals. Is attached. In this OCR, a character discriminating unit 3A having a different configuration is connected to the output side of the character extracting unit 2 instead of the character discriminating unit 3 in FIG. The character discriminating unit 3A receives the character image S2 from the character clipping unit 2, performs character discrimination in the same manner as the character discriminating unit 3, generates a discrimination result S3Aa, and stores contour data when contour tracking is performed. The created contour table S3Ab is created. An image correction unit 8 is connected to the output side of the character extraction unit 2 and the output side of the character determination unit 3A. When the determination result S3Aa of the character determination unit 3A indicates an outline character, the image correction unit 8 corrects the character area S2 so as to be equivalent to a normal character. The character recognition unit 4 is connected to the output side of the image correction unit 8. Other configurations are the same as those in FIG.

FIG. 9 is a block diagram of the image correction unit 8 in FIG. The image correction unit 8 has a switching unit 8a.
The switching unit 8a receives the character area S2 from the character cutout unit 2, receives the determination result S3Aa and the contour table S3Ab from the character determination unit 3A, and if the determination result S3Aa indicates a normal character, the character area S2 is left as it is. The outline table S3Ab and the character image S2 are sent to the outline rectangle detection unit 8b when the judgment result S3Aa indicates an outline character. The outline rectangle detection unit 8b scans the outline table S3Ab and obtains a minimum rectangle surrounding each outline data as an outline rectangle. On the output side of the outline rectangle detection unit 8b, there is provided a correction image creation unit 8c that converts pixels in the character area S2 into pixels in a region surrounded by the outline data in order from upper-level outline data into black pixels or white pixels. It is connected.

Next, processing (1) to OCR in FIG.
(3) will be described. (1) The image storage unit 1 and the character cutout unit 2 perform the same processing as in the first embodiment. (2) Processing in the Character Discrimination Unit 3A FIG. 10 is a diagram showing the outline table S3Ab. The outline table S3Ab includes the number (indicating the detection order) N of the detected normal outline data (1, 2, 3, 3 in FIG. 1),
It is composed of coordinate data D having the origin at the upper left of the character image of the normal outline data, and a rank M described later. The initial value of the order M is 1. The character discriminating unit 3A receives the character area S2 from the character clipping unit 2, performs character discrimination in the same manner as in the first embodiment, and sends the discrimination result S3Aa and the contour table S3Ab shown in FIG. I do.

(3) Processing in the image correction unit 8 The image correction unit 8 receives the character area S2 from the character cutout unit 2, receives the determination result S3Aa and the outline table S3Ab from the character determination unit 3A, and determines the determination result S3Aa. Indicates a normal character, the character area S2 is sent to the character recognition unit 4 as it is. On the other hand, when the determination result S3Aa indicates a blank character, the image correction unit 8 corrects the character in the character area S2 so that the character is the same as a normal character. Next, a method of this correction will be described. Outline rectangle detection unit 8 in image correction unit 8
b, the minimum value Xmin, x of the x coordinate for each contour data
Four elements of a maximum value Xmax of the coordinate, a minimum value Ymin of the y coordinate, and a maximum value Ymax of the y coordinate are obtained.

FIG. 11 is a diagram showing an example of an outline rectangle of the outline data shown in FIG. The contour rectangles in FIG. 11 are compared to rank the contour rectangles. This ranking method compares a certain contour rectangle with another contour rectangle, and when one contour rectangle completely surrounds the other contour rectangle, the enclosed contour data is ranked higher and the The lower contour data is positioned lower, and the rank M of the lower positioned contour data on the contour table S3Ab is lowered by one.
The determination as to whether or not to enclose is made by comparing the above four elements. For example, the first contour data d in FIG.
When comparing 1 with the second contour data d2, the following equation (3) is used.
Is established, it is determined that the second contour data d2 is lower order, and the contour table S of the second contour data d2 is determined.
The rank M on the 3 Ab is lowered by one. First contour data d1 Second contour data d2 Xmin <Xmin Xmax> Xmax Ymin <Ymin Ymax> Ymax (3) The contour rectangle detecting unit 8b performs these ranking operations on all the contour data. Then, the order is determined, and the outline table S3Ab and the character area S2 are sent to the corrected image creating unit 8c. The corrected image creating unit 8c converts the pixels in the character area S2 in the area surrounded by the contour data in order from the upper contour data. This conversion is performed as shown in FIG.
The odd-numbered contour data is converted to black pixels, and the even-numbered contour data is converted to white pixels. FIG.
(A), (b), (c) is a figure which shows the process of an image correction. In the case of the character image shown in FIG.
By converting the region on the character image surrounded by the first-order outline data d1 into black pixels, the image shown in FIG. 12B is obtained. Next, a region surrounded by the second-order contour data d2 and d3 in FIG. 11 is converted into white pixels, whereby a corrected image S8c shown in FIG. 12C is created. The corrected image S8c is sent to the character recognizing unit 4 and subjected to a recognition process. As described above, in the third embodiment, even when a document in which white characters are present is to be recognized, the character determination unit 3A automatically determines whether the character is a normal character or a white character, and determines the character. When the result S3Aa indicates an outline character, the character line is corrected as if it were a normal character composed of only black pixels, and then the recognition process was performed. , The ordinary characters can be recognized as before, and white characters can be correctly recognized.

The present invention is not limited to the above embodiment,
Various modifications are possible. For example, there are the following modifications. (A) In the embodiment, the contour tracing method is used for character discrimination and creation of contour data.
Other methods may be used as long as they can detect the numbers of normal contour data and loop contour data. (B) In the embodiment, the threshold value T for performing character determination
Although HL is set to 2, any other value may be used as long as it is a threshold value that can distinguish between a normal character and an outline character. (C) The binary image ST that is an input signal of the image storage unit 1 may be a signal obtained by binarizing a character reading signal by an image reading device such as a scanner or a facsimile.

[0020]

As described in detail above, according to the first and second aspects of the present invention, even if a document including white characters is to be recognized, the character discrimination unit determines whether the character is a normal character or a white character. Since it is determined whether the character is a character and white characters that are likely to be erroneously read are not read, the recognition accuracy does not decrease. Further, since the recognition result is output as a reject (unread), the operator can easily confirm the recognition result, perform a correction operation, and the like in the processing by the correction terminal in the subsequent stage. According to the third invention, a normal character dictionary and a white character dictionary are prepared in advance, and even if a document in which white characters exist is to be recognized, the character discriminating unit determines whether the character is a normal character. Since it is determined whether the character is a blank character, the dictionary selection unit selects the optimal dictionary according to the determination result of the character determination unit and sends it to the character recognition unit.
White characters can be correctly recognized without lowering the recognition accuracy of normal characters. According to the fourth and fifth aspects of the invention, even when a document in which white characters are present is to be recognized, the character determination unit automatically determines whether the character is a normal character or a white character, and determines that the character is white. In the case of indicating a blank character, the character line is corrected like a normal character composed of only black pixels, and then the recognition process is performed. Therefore, the character recognition unit uses only a conventional normal character recognition dictionary. Normal characters can be recognized as usual, and white characters can be correctly recognized.

[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 diagram illustrating an example of a normal character.

FIG. 3 is a diagram illustrating an example of outline characters.

FIG. 4 is a diagram illustrating a contour tracking method.

FIG. 5 is a diagram showing outline data of a normal character.

FIG. 6 is a diagram showing outline data of outline characters.

FIG. 7 is a configuration diagram of an OCR according to a second embodiment of the present invention.

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

9 is a configuration diagram of an image correction unit 8 in FIG.

FIG. 10 is a diagram showing an outline table S3Ab.

FIG. 11 is a diagram showing a contour rectangle.

FIG. 12 is a diagram illustrating a process of image correction.

[Explanation of symbols]

 T medium 1 Image storage unit 2 Character extraction unit 3 Character discrimination unit 4, 4A character recognition unit 5 Dictionary selection unit 6 Normal character dictionary unit 7 White character dictionary unit 8 Image correction unit 8a Switching unit 8b Outline rectangle detection unit 8c Correction Image creation section

Claims (6)

[Claims] An image storage unit for inputting and storing a binary image image of a medium on which characters are written, and a character extracting unit for extracting a character area from the binary image image stored in the image storage unit And a character recognition unit that recognizes a character in the character region, wherein the character in the cut-out character region has a character line forming the character formed of only black pixels. A character discriminating unit for discriminating whether the character is a normal character or an outline character composed of black pixels only at the edges of the character line is provided.If the judgment result of the character identification unit indicates the outline character, A character recognition device, wherein the character is not recognized. 2. An image storage unit for inputting and storing a binary image image of a medium on which characters are written, and a character extracting unit for extracting a character area from the binary image image stored in the image storage unit The character in the cut-out character area is a normal character in which a character line forming the character is composed of only black pixels or a white character in which only the edge of the character line is composed of black pixels. A character determination unit that determines whether the character is a character, a normal character dictionary unit storing a recognition dictionary suitable for recognition of the normal character, and a white character dictionary unit storing a recognition dictionary suitable for recognition of the white character, A dictionary selection unit that selects the normal character dictionary unit or the outline character dictionary unit based on the determination result of the character determination unit; and recognizes characters from the character region using a recognition dictionary selected by the dictionary selection unit. It has a character recognition unit. Character recognition device. 3. The character discriminating unit includes: a normal contour score obtained by counting pixels constituting a contour outside a black pixel region of a character in the cut-out character region; The number of loop contour points obtained by counting the pixels constituting the inner contour is detected, and the ratio between the normal contour points and the loop contour points is compared with a predetermined threshold value. 3. A structure for determining whether the character is a white character or the white character.
Character recognition device according to the description. 4. An image storage unit for inputting and storing a binary image image of a medium on which characters are written, and a character extracting unit for extracting a character area from the binary image image stored in the image storage unit The character in the cut-out character area is a normal character in which a character line forming the character is composed of only black pixels or a white character in which only the edge of the character line is composed of black pixels. A character discriminating unit for discriminating whether the character is a character, when the discrimination result of the character discriminating unit indicates the normal character
An image correction unit that outputs the characters in the cut-out character area as they are and corrects and outputs the characters in the cut-out character area to the normal characters when the determination result indicates the outlined characters; A character recognition unit that recognizes a character output from the image correction unit. 5. The normal contour data indicating the number of normal contour points obtained by counting the pixels constituting the contour outside the black pixel area of the character in the cut-out character area and the coordinates of the pixel. And the number of loop contour points obtained by counting the pixels constituting the contour inside the black pixel region of the character, and comparing the ratio between the normal contour points and the loop contour points with a predetermined threshold value. 5. The character recognition device according to claim 4, wherein the character is determined to be the normal character or the outline character based on a result, and the normal outline data is output. 6. An image storage unit for inputting and storing a binary image image of a medium on which characters are written, and a character extracting unit for extracting a character area from the binary image image stored in the image storage unit And a character discriminator according to claim 5, wherein when the discrimination result of the character discriminator indicates the normal character,
An image correction unit that outputs the characters in the cut-out character area as they are and corrects and outputs the characters in the cut-out character area to the normal characters when the determination result indicates the outlined characters; A character recognition unit that recognizes a character output from the image correction unit, wherein the image correction unit receives the character region from the character cutout unit, and determines the determination result from the character determination unit and the normal result. When the determination result indicates the normal character, the character area is sent to the character recognition unit, and when the determination result indicates the outline character,
A switching unit that sends the normal outline data and the character area to an outline rectangle detection unit; and a minimum rectangle surrounding the outline is calculated from the normal outline data, and the calculation is performed in an XY coordinate system including orthogonal X and Y axes. As a result of comparing the coordinate values specifying the position and size of each of the rectangles, when a certain rectangle completely surrounds another rectangle, the normal outline data forming this enclosed rectangle is set as the higher order, and the The contour rectangle detecting unit that ranks the respective contour data by positioning the regular contour data forming the smaller rectangle in the lower position, and the regular contour data ranked in the contour rectangle detecting unit and the regular contour data. A character area is received, and pixels in each area surrounded by each normal contour data in the character area are converted to black pixels when the order is odd, and when the order is even, White By performing the process of converting the element from the host in order, and a corrected image creating unit for transmitting a character in the character region is corrected to the normal character to the character recognition unit,
A character recognition device comprising: