JP2000181989A - Character recognition device, method therefor and computer readable recording medium recording the method as program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a character recognition device for finding a horizontally written character string included in vertically written character rows in character images and correctly recognizing characters. SOLUTION: A binary picture input part 101 inputs the document images of a recognition object as binary pictures and a character constituting element detection part 102 detects character constituting elements from the binary pictures. A character row extraction part 103 connects the character constituting elements and extracts a character row. A character row direction judgement part 104 judges the direction, an extraction part 105 extracts a character area from the character row and a character recognition part 107 collates it with a recognition dictionary 106 and recognizes the character. A horizontally written character string judgement part 108 judges the presence/absence of the possibility of the horizontally written character string based on the recognized character. At the time of judging that it is present, a horizontally written character string division recognition part 110 divides the character images in a horizontal direction and recognizes new character images by collating them with the recognition dictionary 106. A recognized result output part 111 selects the one of high similarities of recognized results.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to a character recognition apparatus and method for recognizing characters in a document image such as a mail or a form and converting the characters into a character code.

[0002]

2. Description of the Related Art An optical character recognition device (OCR) is effective in recognizing a destination address of a mail. Generally O
In CR, a character line is first extracted from an optically read character image, and each extracted character line is separated in the line direction to perform character extraction, and character recognition is performed for each extracted character. . Therefore, a horizontal character string in a vertical character line,
For example, when "40" is described, the character is extracted as one character, so that the character cannot be correctly recognized.

In order to solve such a problem, as a first device, a character line recognizing device described in Japanese Patent Application Laid-Open No. Hei 5-81463 is used. A recognition result when a horizontal character string is recognized as one kanji character is prepared in a table in advance, and when a recognition result of a character in a vertical character line includes a character described in the table, the recognition result is equivalent to the character. By converting to a numeric string, a horizontal numeric string in a vertical character line is recognized.

As a second device, Japanese Patent Application Laid-Open No.
There is a character recognition device described in Japanese Patent No. 719. With this device, wide characters in a vertical writing character line and characters with poor recognition result evaluation values are extracted as horizontal characters, and horizontal character extraction based on projection is performed, and character recognition is performed again. Then, the horizontal character string in the vertical character line is recognized.

[0005]

By the way, the first device is used only for recognizing a destination address of a postal matter. Characters described in a table in advance are characters to be converted into a numeric string. It must be judged by using an address dictionary together. Further, in the second device, when characters in a horizontal writing character string in a vertical writing character line are in contact with each other, character extraction cannot be performed correctly because character extraction is performed based on projection. Cannot recognize characters correctly.

Therefore, the present invention provides a method for accurately recognizing a horizontally written character string even when a horizontally written character string is included in a vertically written character line and the horizontally written characters are in contact with each other. It is an object of the present invention to provide a character recognition device and a method thereof.

[0007]

In order to solve the above-mentioned problems, the present invention relates to a character component detecting means for detecting a character component from an input binary character image, and concatenating the detected character components. A character line extracting unit for extracting a character line, a vertical / horizontal determining unit for determining whether the extracted character line is vertical or horizontal according to the longitudinal direction of the character line, and a character image as one character in the character line. Character image area detecting means for detecting a first character image area constituting the first character image area; first character recognizing means for recognizing a character as one character from the detected first character image area; If it is determined that the character is written, the horizontal character string determination unit determines that the character image in the first character image area is a horizontal character string based on the recognition result of the character recognition by the first character recognition unit; means A second character recognition unit that divides the first character image region into a plurality of second character image regions in a horizontal direction when the character image is determined to be a horizontal writing character string, and recognizes a character from a character image in each second character image region; There is provided a recognition result selecting means for selecting any of the character recognized by the first character recognizing means and the character string recognized by the second character recognizing means as a recognition result.

[0008]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the character recognition device according to the present invention will be described with reference to the drawings. (Embodiment) FIG. 1 is a configuration diagram of an embodiment of a character recognition device according to the present invention. This character recognition device includes a binary image input unit 101, a character component detection unit 102, a character line extraction unit 103, a character line direction determination unit 104, a character extraction unit 105, a recognition dictionary 106, a character recognition Part 107
, A horizontal character string determination unit 108, a loose character component integration unit 109, a horizontal character string division recognition unit 110, and a recognition result output unit 111.

A binary image input unit 101 has a scanner or the like and inputs a document image to be recognized as a binary image. FIG. 2 is a diagram illustrating an input binary image. Binary image 2
01 indicates a character component (including a character and a character element) as a black pixel and a background portion as a white pixel. The binary image 2
01 is input in a direction in which characters can be normally read when FIG. 2 is viewed from the front.

The character component detection unit 102 has a storage area and has a binary image 20 input by the binary image input unit 101.
A character component that is a block of black pixels continuing from 1 is detected and stored. FIG. 3 shows character components 301, 302, 303,... Detected from the binary image 201. At this time, the character component detection unit 102 integrates character components having large overlapping circumscribed rectangles of the character components. The “fu” 305 is integrated. The character component detecting unit 102 notifies the character line extracting unit 103 and the character extracting unit 105 of the coordinate values of the circumscribed rectangle of the character component.

In FIG. 3, the circumscribed rectangle of the character component is not in contact with the character component, but this is caused by a limitation in drawing, and in fact, it is in contact with the character component. The same applies to the following drawings. The character line extraction unit 103 includes a character component search unit 102
When the notification of the coordinate values of the circumscribed rectangles of the character components is received from, the individual character components 301, 302,.
Is integrated step by step to extract character lines. Binary image 201
As shown in FIG. 4, character lines 401 and 402 are extracted. The coordinate values of the circumscribed rectangles of the extracted character lines 401 and 402 are notified to the character line direction determination unit 104 and the character extraction unit 105.

The character line direction determining unit 104 determines the longitudinal direction of the character lines 401 and 402 as the direction of the character line based on the notified coordinate values. Character lines 401 and 402 shown in FIG.
Indicates the vertical direction, and is determined to be vertical writing. The character line direction determining unit 104 notifies the character extracting unit 105 of the direction of the character line. When the character extracting unit 105 receives the notification of the direction (vertical writing) of the character line from the character line direction determining unit 104, the character extracting unit 105 determines the coordinate value of the circumscribed rectangle of the character component notified from the character component detecting unit 102, From the coordinate values of the circumscribed rectangle of the character line notified from the extraction unit 103, the coordinate values of the character area forming one character in each of the character lines 401 and 402 are sequentially notified to the character recognition unit 107.

The determination of the character area constituting one character is performed based on the following rules. (1) Based on the arrangement of the character components in the character line, FIG.
The reference character height and character spacing are determined as shown in FIG. (A) The reference character height H is a maximum value of the height of a character component that does not exceed the character line width W.

(B) The reference character spacing D is the minimum value of the character component spacing in which the spacing between vertically adjacent character components exceeds the reference character height H. (2) Based on the reference character height H and the reference character spacing D, adjacent character components are integrated to determine the area of each character. (A) The character components adjacent to the left and right are integrated.

(B) When character components are vertically adjacent to each other, they are integrated within a range not exceeding the reference character interval D. FIG. 5B shows a character image extracted from the vertically written character line 401 by the character extracting unit 105 based on the above-described rule. Postal code mark "@" 501, digit string "588" 5
02, vertical hyphen "|" 503, numeric string "511"
Numeral string “511” 505 out of “1” 504,.
Numeric string “40” 506,..., Numeric string “14” 507
Is extracted as a one-character image.

In the case of a horizontally written character line, the character area is determined in the same manner by changing the length and width. The recognition dictionary 106 includes a normal recognition dictionary portion in which a set of a standard feature that is a feature of a standard character image of a character and its character code is registered, a standard feature of a two-digit numeric image, and a character code of a two-digit numeric. And a two-part number cluster dictionary part in which a set of numbers is registered. The two-digit number image is a two-digit number image printed in a plurality of types (four types in this case) as shown in FIG. 6, and its standard feature is extracted from an image of all combinations of two-digit numbers. The feature is analyzed in a cluster, and the average feature of each cluster is obtained.

Upon receiving a notification of the coordinate value of the character area to be recognized from the character extracting section 105, the character recognizing section 107 extracts a feature from the character components of the character area stored in the character component detecting section 102. Then, the similarity between the standard feature of the recognition dictionary 106 and the extracted feature is calculated, and the character code stored as a pair with the standard feature having the highest similarity is used as the recognition result.

FIG. 7 shows the recognition result of the character image in the vertically written character line 401. The character recognition unit 107 notifies the recognition result to the horizontally written character string determination unit 108 and the recognition result output unit 111. Note that the recognition result output unit 111 is notified of the similarity in addition to the character code. The horizontal writing character string determination unit 108 determines the possibility of a horizontal writing character string in the vertical writing character line 401 based on the character code notified from the character recognition unit 107.

The following rules are applied as criteria for determining the possibility of a horizontally written character string. Rule (a) Two-consecutive digits recognized in the two-consecutive-number cluster dictionary rule (b) Characters recognized other than digits following various postal code marks or characters recognized by vertical or horizontal hyphens Rule (c ) Characters Recognized in Alphabet In addition to the above rules, the horizontal writing character string determination unit 108 stores a candidate character table (not shown) that is easily misrecognized, for example, “00” and “∞”. In addition, a character recognized as a character described in the table may be determined as a horizontally written character string.

The judgment result of the possibility of the horizontal writing character string judging section 108 is shown with a circle next to the recognition result shown in FIG. Since the character image of the numeral string “588” 502 is recognized as the alphabet “m” other than the number following the character recognized as the postal code mark, the rule (b) is applied. In the character image of the numeral string “511” 505, the rule (b) is applied because the character following the vertical hyphen is recognized as the kanji “familiar”. Numeric string "40" 5
Since the character image 06 is recognized as a two-digit number with the numeral “40”, the rule (a) is applied. Since the character image of the numeral string “14” 507 is recognized as the alphabet “M”, the rule (c) is applied.

When it is determined that there is a possibility of a horizontally written character string, the horizontally written character string determination unit 108 compares the coordinate value of the area of the character image and the determination rule with the free character component integration unit 1.
09 and the horizontal writing character string division recognition unit 110 is notified. When the document image input from the binary image input unit 101 is a horizontally written character string as shown in FIG.
01 is determined as a horizontally written character string by the character line direction determination unit 104, and the character is recognized by the character recognition unit 107 as shown in FIG. In this case, the horizontal writing character string determination unit 108 determines the possibility of re-division of the character image by applying the rules (a) and (b).

A character image determined to have a possibility of re-division in combination with the recognition result shown in FIG. 9 is marked with a circle. The character images 802, 803, and 804 in the character string 801 correspond to the rules (b), (a), and (b), respectively, and it is determined that there is a possibility of re-division, and the coordinate values of the area of the character image are written horizontally. The character string division recognition unit 110 is notified. Upon receiving notification from the horizontal character string determination unit 108 that there is a possibility of a horizontal character string, the loose character component integrating unit 109 moves the character line extraction unit 103 to a distance less than or equal to a predetermined value from the character image area.
Searches the storage area of the character component detection unit 102 for any loose character components that are not included in the character line 401. If there is a loose character component, it is included in the area of the character image, and the horizontal character string division recognition unit 110 is notified of the coordinate value of the area.

FIG. 10 shows the free character component integrating unit 109.
FIG. 9 is a diagram for explaining that a character component 1001 separated from the image is integrated into a character image 505. Character line extraction unit 103
Then, when the character line 401 is extracted, one character image 50
It is generally possible to exclude character components 1001 that should be included in 5. The free character component integration unit 109
Such a disadvantage is solved.

When the horizontal writing character string division recognizing section 110 receives a notification from the horizontal writing character string determination section 108 that there is a possibility of a horizontal writing character string, the rule of the notified character image is used as a basis for the determination. And divide horizontally. The feature is extracted from the new divided character image, the similarity with the standard feature of the recognition dictionary 106 is calculated, and the character code of the character having the highest similarity is notified to the recognition result output unit 111. At this time, the similarity value is also notified.

For example, in the case of a character image of the numeral string "588" 502 recognized by the character recognition unit 107 as the alphabet "m", the horizontal character string division recognition unit 110 sends a rule (b) from the horizontal character string determination unit. ) Has been applied. Since the character immediately before the character image is a postal code mark, the character image of the numeral string "588" 502 is horizontally divided into three parts by projection. An image 1101 (FIG. 11B) in which white pixels sandwiched between upper and lower black pixels of the numeric string “588” 502 shown in FIG. 11A are filled with black pixels, and a vertical projection operation is performed. Performed projected image 1102 (FIG. 1)
1 (c)). The projected valleys 1103 and 1104 are detected, the image is divided by the valleys, and a new character image 110 is detected.
5, 1106 and 1107 are obtained (FIG. 11D). As a result, even if the character images 1105, 1106, and 1107 are touching character strings, the character images can be separated. For each of the obtained character images 1105, 1106, and 1107, the feature is extracted, the similarity with the standard feature of the recognition dictionary 106 is calculated, and the character code of the character with the highest similarity is notified to the recognition result output unit 111. .

The horizontal character string division recognition unit 110
In the case of the two-digit number “40” 506, since the rule (a) has been notified from the horizontal character string determination unit 108,
The character image of the serial number “40” 506 is equally divided in the horizontal direction to obtain new character images 1201 and 1202 (FIG. 1).
2). The feature amount is extracted from each of the character images 1201 and 1202, is compared with the recognition dictionary 106, and the recognition result is notified to the recognition result output unit 111.

In the example of the character image shown in FIGS. 11 to 12, the number of divisions to be divided in the horizontal direction is determined in advance according to a rule or the like on which the determination that the character string is a horizontal writing character string is made in advance. However, the following method can be applied as another method of dividing a horizontally written character string. (A) When the character image is composed of a plurality of character components, the reference width is obtained by adding the width of the character component having the smallest width and the interval between adjacent character components. The number of characters is estimated, a projection operation is performed, and the image is divided into character images having the estimated number of characters. (B) When the character image is composed of a single character component, the number of characters is estimated based on half the height of the character component, and divided into a new character image as in (a) above. I do.

The horizontal character string division recognition unit 110 has been notified of the rule (b) from the horizontal character string determination unit 108. If the preceding character is a vertical hyphen or, as shown in FIG. When the rule (b) is notified from the column determining unit 108, the preceding character is a vertical writing hyphen, and the area of the integrated character image 505 is notified from the loose character component integrating unit 109,
By applying the above method, new character images 1301, 130
2, 1303 and 1304 are obtained. Character image 13 obtained
01 to 1304, respectively, and a recognition result output unit 11
Notify 1.

FIG. 14 shows the result of character recognition by the horizontally written character string division recognition unit 110 in the character line 401 and the distance (similarity) thereof. For reference, the character recognition unit 1
The recognition result at 07 is also shown. Note that, even in the case of the horizontally written character string 801 as shown in FIG. 8, the horizontally written character string division recognition unit 110 performs the following processing on the character image determined to be re-segmentable by the horizontally written character string determination unit 108.
The number of characters is determined or estimated based on a predetermined rule, and a contact character or the like is divided to recognize a character from a new character image.

The recognition result output unit 111 outputs the similarity derived from the recognition result of the character image determined to be a horizontal writing character string by the horizontal writing character determination unit 108 by the character recognition unit 107 and the similarity obtained by the horizontal writing character string division recognition unit 110. The value of the similarity with which the recognition result was derived is compared, and a recognition result with a higher similarity is selected as a recognized character of the character image and displayed on the display screen.

Since the horizontal character string division recognition unit 110 recognizes a plurality of horizontal character strings, the similarity of each character is calculated. The character recognizing unit 107 normally recognizes the character as a single character, so the degree of similarity is one. Therefore, when comparing the similarities, the similarity in the character recognition unit 107 and the horizontal writing character string division recognition unit 110 are compared.
By comparing the lowest similarity among multiple similarities in
The higher similarity may be selected, or the comparison may be performed using the average value of a plurality of similarities in the horizontally written character string division recognition unit 110.

In FIG. 14, since the similarity is shown as a distance, a high similarity means that the distance is short. Next, the operation of the present embodiment will be described with reference to FIGS.
This will be described with reference to the flowchart of FIG. First, the binary image input unit 101 inputs a document image to be recognized as a binary image (S1502).

The character component detecting section 102 detects a character component (black pixel) from the binary image (S1504), and the character line extracting section 103 connects the detected black pixels to form a character line (character line). Is extracted (S150).
6). The character line direction determination unit 104 determines whether or not the character line is written horizontally by checking whether the longitudinal direction is the vertical direction or the horizontal direction based on the coordinate values of the extracted character line (S1508). If not, the process moves to S1702.

When the result is affirmative, the character extracting unit 105 extracts a character image (coordinates of a circumscribed rectangle of the character image) as one character from the extracted character line (S1510). The character recognizing unit 107 determines whether there is any unrecognized character in the extracted character line (S1512). If not, the process ends. If there is, the character in the character image area is recognized (S1512). S1
514).

The horizontally written character string determination unit 108 determines whether or not the recognized character is a two-digit number (S151).
6). If not, the horizontal writing character string division recognition unit 110
The character image is equally divided in the left-right direction (S1518), and the characters are re-recognized from the two new character images (S1518).
1520). The recognition result output unit 111 includes the character recognition unit 10
7 and the horizontal character string division recognition unit 110
Compare the two similarities with the character string recognized in the above, select a character or a character string with a high similarity, display it on the display screen (S1
522), and return to S1512.

If it is determined in S1516 that the character string is not, in S1602, the horizontal character string determination unit 108
Determines whether the recognized previous character is a postal code mark (1602). When the result is affirmative, the horizontally written character string division recognition unit 110 divides the area of the next character image recognized by the postal code mark into three parts by the projection operation (S160).
4). Characters are re-recognized from the three new divided character images (S1606).

The recognition result output unit 111 includes a character recognition unit 107 and a horizontally written character string division recognition unit 110 as in S1522.
(S1).
608), and return to S1512. When determining in S1602 that the character is not, the horizontal writing character string determination unit 108 determines whether or not the preceding character is a vertical writing hyphen (S161).
0). If no, the process returns to S1512.

When the result is affirmative, the horizontal writing character string division recognition unit 1
In step S1612, the character 10 estimates the number of characters of the character image next to the character recognized as a hyphen based on the reference width or half the height of the character component, and divides the character image (S1612).
In S1508, when the character line direction determination unit 104 determines that the character line is horizontal writing, the character extraction unit 105 extracts a character image as one character (S1702).

The character recognizing unit 107 determines whether or not there is a character (S1704), and if not, ends the processing. If there is a character, the feature is extracted and the character is recognized by collating with the recognition dictionary 106 (S1706). The horizontal writing character string determination unit 108 determines whether the recognized character is a two-digit number (S1708). If it is not a two-digit number, S1
Return to 704.

In the case of two consecutive numbers, the horizontally written character string division recognition unit 110 divides the character image equally into two in the left-right direction, extracts the characteristics of each character image, compares it with the recognition dictionary 106, and compares it with the recognition dictionary 106. Is recognized (S1712). The recognition result output unit 111 selects a character or a character string having a high degree of similarity from among the characters or character strings recognized by the character recognition unit 107 and the horizontally written character string division recognition unit 110 (S171).
4) Return to S1704.

It should be noted that if the determination is affirmative in S1604 and S1610, the free character component integrating unit 1
09 checks whether there is a character component separated within a predetermined distance from the character line outside the left and right (horizontal direction) of the character image to be recognized. Integrate. The horizontal writing character string division recognition unit 110 performs a projection operation on the integrated character image to divide the character image.

Although the character recognition apparatus according to the present invention is shown in FIG. 1 as a configuration diagram in the above-described embodiment, a computer-readable recording program such as a floppy disk can be used to execute a program that performs the functions of each component. By recording the data on a medium and transferring the data, or transferring the data via a communication circuit, it can be easily implemented by another independent computer system.

[0043]

As described above, the character recognition device according to the present invention connects a character component detecting means for detecting a character component from an input binary character image with the detected character component. A character line extracting means for extracting a character line,
Vertical / horizontal determining means for determining whether the extracted character line is vertical or horizontal according to the longitudinal direction of the character line, and a character for detecting a first character image area constituting a character image as one character in the character line An image area detecting unit, and a first character recognition unit that recognizes a character as one character from the detected first character image area
When a character line is determined to be vertical writing by the character recognition means and the vertical / horizontal determination means, the character image in the first character image area is defined as a horizontal writing character string based on the recognition result of character recognition by the first character recognition means. When the horizontal writing character string determining means determines that the character string is a horizontal writing character string, the first character image area is divided into a plurality of second character image areas in the horizontal direction. A second character recognizing means for recognizing a character from the character image in the area, and a recognition result selection for selecting any of a character recognized by the first character recognizing means and a character string recognized by the second character recognizing means as a recognition result Means. With such a configuration, characters in a horizontal writing character string included in a vertical writing character line can be correctly recognized.

When the character recognized by the first character recognizing means is a postal code mark, the horizontal character string judging means may determine the character recognized by the first character recognizing means following the postal code mark. When a character other than a number is recognized, a postal code determining unit that determines a character image of the first character image area recognized as a character other than the number as a horizontally written character string,
The second character recognizing means has a first division unit for horizontally dividing the determined first character image area into three when the postal code determination unit determines that the character string is a horizontal writing character string. With such a configuration, it is possible to correctly recognize a three-digit horizontal digit string following the postal code mark.

When the character recognized by the first character recognizing means is a vertical hyphen, the horizontal character string judging means recognizes the character recognized by the first character recognizing means following the hyphen. When it is recognized other than numbers,
A hyphen determining unit that determines a character image of the first character image area recognized as a character other than the numeral as a horizontally written character string, wherein the second character recognition unit determines that the character image is a horizontally written character string by the hyphen determining unit; And a second division unit for dividing the determined first character image area in the horizontal direction. With such a configuration, it is possible to correctly recognize a horizontal numeral string following a hyphen.

Further, the horizontal character string determining means has an alphabet determining part for determining a character image in the first character image area as a horizontal character string when the character recognized by the first character recognizing means is an alphabet. And
With such a configuration, in a vertically written character line, a character image recognized as an alphabet having a low appearance frequency can be correctly recognized.

Further, the second character recognizing means, when a character component not included in the first character image area in the lateral direction of the first character image area is within a predetermined distance, the second character recognition means removes the character component from the first character image area. It has an integration unit for integrating into one character image area. With such a configuration, it is possible to correct a character line extraction error that is likely to occur in general, and perform character recognition correctly.

Further, the second character recognizing means includes the first character recognizing means.
The area surrounded by the character components included in the character image region is converted into the same elements as the character components, vertical projection processing is performed, and the first character image region is divided at the valleys of the projected shape. It has a valley division. With such a configuration, it is possible to obtain a character image obtained by dividing a horizontally written character string with high accuracy, so that character recognition can be performed correctly.

Further, the first character recognizing means has a recognition dictionary for recognizing one character from the character image in the first image area, and the two-digit number is combined with the recognition dictionary. The two-character string is registered, and the horizontal character string determination means determines that the character image in the first character image area is a horizontal character string when the first character recognition means recognizes the character as a two-character string. It has a concatenated-number determining unit, and the second character recognizing means has a third dividing unit that divides the first image area determined as a horizontally written character string by the two-concatenated-number determining unit into two equal parts in the horizontal direction. I have. With such a configuration,
Even if the contacted two-digit number is included in the vertical character line, the character can be correctly recognized.

[0050] The recognition result selecting means may include the first
There is provided a selection unit that selects the one with the higher similarity between the character recognized by the character recognition unit and the character string recognized by the second character recognition unit. With such a configuration,
A more reliable character can be determined as a recognition result from the recognized characters. In addition, the character recognition method according to the present invention includes a character component detecting step of detecting a character component from an input binary character image, and a character line extracting step of connecting the detected character components to extract a character line. Step, a vertical / horizontal determination step of determining whether the extracted character line is vertical or horizontal according to the longitudinal direction of the character line, and a first character image area forming a character image as one character in the character line. Detecting a character image area to be detected, and performing first character recognition as one character from the detected first character image area
In the character recognition step, when the character line is determined to be vertical writing by the vertical / horizontal determination means, the character image in the first character image area is defined as a horizontal character string based on the result of character recognition in the first character recognition step. Determining a horizontal writing character string in the horizontal writing character string determining step; and determining a first character image area in a plurality of second horizontal writing character strings when the horizontal writing character string is determined in the horizontal writing character string determining step.
A second character recognition step of dividing the character image into character image areas and recognizing characters from the character images of the second character image areas; a character recognized by the first character recognition means and a character string recognized by the second character recognition step; And a recognition result selecting step of selecting any of the above as a recognition result. With such a configuration, characters in a horizontal writing character string included in a vertical writing character line can be correctly recognized.

Further, the present invention is a computer-readable recording medium, comprising a character component detecting step of detecting a character component from an input binary character image, and linking the detected character components. A character line extraction step of extracting a character line, a vertical / horizontal determination step of determining whether the extracted character line is vertical writing or horizontal writing according to the longitudinal direction of the character line, and a character image as one character in the character line. A character image region detecting step of detecting a first character image region to be configured; a first character recognition step of recognizing a character as one character from the detected first character image region; If it is determined that the first
Based on the recognition result of the character recognition in the character recognition step, the character image in the first character image area is determined to be a horizontal character string, and the horizontal character string determination step is determined to be a horizontal character string. At this time, the first character image area is divided into a plurality of horizontal second character image areas,
A second character recognition step of recognizing a character from a character image in each second character image area; and a character recognized by the first character recognition means or a character string recognized by the second character recognition step as a recognition result. And a computer-readable recording medium storing a method having a selecting step of selecting a recognition result. With such a configuration, such a function can be exhibited by applying to a character recognition device that does not have a function of correctly recognizing characters in a horizontal writing character string included in a vertical writing character line.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of an embodiment of a character recognition device according to the present invention.

FIG. 2 is a diagram illustrating a binary document image to be recognized input by a binary image input unit according to the embodiment.

FIG. 3 is a diagram illustrating character components detected by a character component detection unit according to the embodiment.

FIG. 4 is a diagram showing character lines extracted by a character line extraction unit according to the embodiment.

FIG. 5A is a diagram illustrating a rule applied by a character extracting unit according to the embodiment. FIG. 4B is a diagram illustrating a region of a character image extracted by the character extraction unit according to the embodiment.

FIG. 6 is a diagram illustrating an example of a two-digit numeric image registered in a two-digit numeric cluster dictionary part of the recognition dictionary according to the embodiment.

FIG. 7 is a diagram illustrating a character (recognition result) recognized from the character line 401 by the character recognition unit according to the embodiment and a character determined to have a horizontal writing character string possibility by the horizontal writing character string determination unit.

FIG. 8 is a diagram illustrating an area of a character image extracted by a character extracting unit when an input document image is a horizontally written character string in the embodiment.

FIG. 9 is a diagram illustrating a result of recognition of the horizontal writing character string shown in FIG. 8 by the character recognition unit and a character determined to be re-dividable by the horizontal writing character string determination unit.

FIG. 10 is a diagram illustrating integration of loose character components in the loose character component integration unit according to the embodiment.

FIG. 11 is a diagram illustrating a state in which a character image is divided by a projection operation in a horizontally written character string division recognition unit according to the embodiment.

FIG. 12 is a diagram illustrating division of a two-part numeric image in a horizontally written character string division recognition unit according to the embodiment.

FIG. 13 is a diagram for explaining division of a character image by a horizontal writing character string division recognition unit for a character image integrated by a loose character component integration unit according to the embodiment.

FIG. 14 is a diagram showing a recognition result in a horizontally written character string division recognition unit of the embodiment.

FIG. 15 is a flowchart illustrating the operation of the embodiment.

FIG. 16 is a flowchart illustrating an operation of the embodiment.

FIG. 17 is a flowchart illustrating an operation of the embodiment.

[Explanation of symbols]

 Reference Signs List 101 Binary image input unit 102 Character component detection unit 103 Character line extraction unit 104 Character line direction determination unit 105 Character extraction unit 106 Recognition dictionary 107 Character recognition unit 108 Horizontal character string determination unit 109 Free character component integration unit 110 Horizontal character Column division recognition unit 111 Recognition result output unit

 ──────────────────────────────────────────────────の Continuing on the front page (72) Inventor Ichiro Nakao 1006 Kadoma Kadoma, Osaka Prefecture Inside Matsushita Electric Industrial Co., Ltd. F term (reference) 5B029 AA01 BB02 CC20 CC28 EE08 5B064 AA01 AA04 AB13 AB16 AB19 CA08 DC08 DC09 DC18 EA26 EA36

Claims (10)

[Claims] 1. A character component detecting means for detecting a character component from an input binary character image, a character line extracting means for connecting the detected character components and extracting a character line, Vertical / horizontal determining means for determining whether the character line is vertical or horizontal according to the longitudinal direction of the character line, and a first character image constituting one character in the character line
A character image area detecting means for detecting a character image area; a first character recognizing means for recognizing a character as one character from the detected first character image area; If
A horizontally written character string determining unit that determines a character image in the first character image area as a horizontally written character string based on the recognition result of the character recognition by the first character recognition unit, and a horizontally written character string that is determined by the horizontally written character string determining unit. The first character image area is divided into a plurality of second character image areas in the horizontal direction, and a second character recognizing means for recognizing characters from the character images in each of the second character image areas; A character recognition device comprising: a recognition result selecting unit that selects, as a recognition result, one of the character recognized by the character recognition unit and the character string recognized by the second character recognition unit. 2. The method according to claim 1, wherein the character string recognized by the first character recognizing means is a postal code mark when the character recognized by the first character recognizing means is a postal code mark. Has a zip code determining unit that determines a character image of the first character image area recognized as a character other than the number as a horizontally written character string when the character is recognized as a character other than the number. When the number judgment unit judges that the character string is a horizontal writing character string,
A first division of the determined first character image area into three in the horizontal direction
The character recognition device according to claim 1, further comprising a dividing unit. 3. When the character recognized by the first character recognizing means is a vertical hyphen, the horizontally-written character string determining means determines whether the character recognized by the first character recognizing means following the hyphen is a hyphen. When a character image other than a number is recognized, the character image in the first character image area recognized as a character other than the number has a hyphen determination unit that determines a horizontal writing character string, and the second character recognition unit includes the hyphen determination unit. Is determined as a horizontal character string,
2. The character recognition device according to claim 1, further comprising a second division unit that divides the determined first character image area in the horizontal direction. 4. A horizontal character string determination unit, comprising: an alphabet determination unit for determining a character image in a first character image area as a horizontal character string when a character recognized by the first character recognition unit is an alphabet. The character recognition device according to claim 1, wherein: 5. The character recognition device according to claim 1, wherein a character component not included in the first character image area in a lateral direction of the first character image area is within a predetermined distance. 2. The character recognition device according to claim 1, further comprising an integration unit that integrates the image into one character image area. 6. The second character recognizing means converts an area surrounded by character components included in the first character image area into the same element as a character component, and performs a vertical projection process. The character recognition device according to claim 2, further comprising a valley dividing unit that divides the first character image area at a position of a valley in a projected shape. 7. The first character recognizing means has a recognition dictionary for recognizing one character from a character image in a first image area, wherein the recognition dictionary combines all two-digit numbers. When two consecutive numbers are registered, and the horizontal writing character string determining means is recognized as two consecutive numbers by the first character recognizing means,
The character image in the first character image area is a horizontally written character string; and the second character recognition means is a first character image determined to be a horizontally written character string by the two-character string determination unit.
2. The character recognition device according to claim 1, further comprising a third dividing unit that divides the image area into two in the horizontal direction. 8. The recognition result selecting unit includes a selecting unit that selects a character having a higher similarity between the character recognized by the first character recognizing unit and the character string recognized by the second character recognizing unit. The character recognition device according to any one of claims 1 to 7, wherein: 9. A character component detecting step of detecting a character component from the input binary character image, a character line extracting step of connecting the detected character components and extracting a character line, A vertical / horizontal determination step of determining whether a character line is vertical or horizontal according to the longitudinal direction of the character line; and a first image forming a character image as one character in the character line.
A character image region detecting step of detecting a character image region; a first character recognition step of recognizing a character as one character from the detected first character image region; and the vertical / horizontal determining means determines that the character line is vertical writing If
A horizontally written character string determining step of determining the character image in the first character image area as a horizontally written character string based on the recognition result of the character recognition in the first character recognition step; and determining the horizontally written character string in the horizontally written character string determining step. The first character image area is divided into a plurality of second character image areas in the horizontal direction, and a second character recognition step of recognizing characters from the character images in the second character image areas; A character recognition method comprising: a recognition result selecting step of selecting, as a recognition result, one of the character recognized in step (1) and the character string recognized in the second character recognition step. 10. A computer-readable recording medium, comprising: a character component detecting step of detecting a character component from an input binary character image; and connecting the detected character components to extract a character line. A character line extracting step, a vertical / horizontal determining step of determining whether the extracted character line is vertical writing or horizontal writing in accordance with the longitudinal direction of the character line; and a first image forming a character image as one character in the character line.
A character image region detecting step of detecting a character image region; a first character recognition step of recognizing a character as one character from the detected first character image region; and the vertical / horizontal determining means determines that the character line is vertical writing If
A horizontally written character string determining step of determining the character image in the first character image area as a horizontally written character string based on the recognition result of the character recognition in the first character recognition step; and determining the horizontally written character string in the horizontally written character string determining step. The first character image area is divided into a plurality of second character image areas in the horizontal direction, and a second character recognition step of recognizing characters from the character images in the second character image areas; A computer-readable recording medium which records a method having a recognition result selecting step of selecting, as a recognition result, one of a character recognized in step (1) and a character string recognized in the second character recognition step.