JPH10154208A - Character recognition device, and storage medium, which computer storing program for functioning computer as character recognition device can read - Google Patents

Abstract

PROBLEM TO BE SOLVED: To precisely detect a character rectangle which is possibly erroneous by judging the character rectangle whose character width is longer than character height based on the reliability of a recognized result, a noticed character and character information before and after it. SOLUTION: An image picture is inputted from a picture input part 101 and a character recognition processing part 104 executes a character recognition processing. A character re-segmenting judgement part 107 assumes the character whose character width is longer than character height as a contact character, and judges the re-segmenting to be necessary when a system applies to one of following three items. (1) The reliability of the noticed character by a reliability calculation part 106 is not more than a threshold, either character before or behind the noticed character is alphanumeric and the noticed character is not the alphanumeric. (2) Information of the part of speech on the noticed character is a non- registered word when the reliability of the noticed character is not more than the threshold, either character before or behind the noticed character is alphanumeric and the noticed character is also the alphanumeric. (3) The characters before and behind are not the non- registered words and the relations on the heights of the noticed character and the characters before and behind are the same.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is directed to a Japanese document. In order to improve the character recognition rate, the present invention detects a contact character which is assumed to be in contact with two characters and erroneously cut out as one character. The present invention relates to a character recognizing device for determining whether or not the character can be re-cut out, re-processing, and the like, and a storage medium storing the character recognizing process.

[0002]

2. Description of the Related Art In recent years, Japanese documents have been increasingly occupied by alphanumeric strings and numeric strings. At the same time, alphanumeric fonts and Japanese fonts have been used together. In addition, the number of fonts has also become diverse.

[0003] Under such circumstances, the pitch between characters in a document is not always constant, and in a copy document, there are many alphanumeric contact characters generated during copying.

For this reason, when character recognition is performed, the character rectangle obtained once is erroneously used in order to accurately extract the character rectangle of half-width characters such as alphanumeric characters and full-width characters of the Japanese part. A method of detecting a possible object and performing re-cutting has been used.

[0005] For example, as a reference technical document relating to such a technique, Japanese Patent Laid-Open No. Hei 4-18881 discloses a "character reading device" that includes a provisional character which is not determined to be a full-width character based on character size position information based on alphanumeric characters. It is disclosed that a character is classified into categories, and a character that could not be classified is regarded as a contact character to perform a division process, and that the validity of a comprehensive character / separated character is determined by a linguistic rule.

Japanese Patent Application Laid-Open No. 4-184584 discloses a "character recognition method and apparatus" in which a character rectangle is divided / re-recognized for a character whose recognition result is not certain.

[0007]

However, the prior art as described above has the following problems.

In Japanese Patent Application Laid-Open No. Hei 4-18881, it is determined whether or not alphanumeric characters are present from information before recognition processing such as character size. However, character re-cutout processing is not performed on the result after recognition. As a result, characters that may be incorrect may not be accurately detected. Further, in Japanese Patent Application Laid-Open No. 4-184584, evaluation and comparison of two recognition results are not performed, so that an inappropriate result may be output as a character.

The present invention has been made in view of the above, and has as its first object to realize accurate detection of a possibly incorrect character rectangle in order to perform re-cutting.

It is a second object of the present invention to minimize an error in clipping by preventing an inappropriate result from being output as a character.

[0011]

According to a first aspect of the present invention, there is provided a character recognition apparatus, comprising: extracting a character string and a character rectangle from input character image information; Recognizing, obtaining a reliability indicating the likelihood of the recognition result, further re-cutting and recognizing the character rectangle according to the reliability, comparing and evaluating the first recognition result and the re-cutting recognition result, In a character recognition device that selects and outputs a likely result, for a character rectangle in which the number of rectangles constituting a character is one and the character width is longer than the character height, the reliability of the recognition result and the target character and the character before and after the target character And character re-cutting determination means for determining whether or not re-cut-out processing for separating characters is to be executed based on the character information.

That is, a character in which the number of rectangles constituting a character is one and whose character width exceeds the character height is two characters.
Assuming that the contact character is a contact character candidate that is assumed to be cut out as one character upon contact, the reliability of the target character is equal to or less than the threshold value, one of the characters before and after the target character is alphanumeric, and the target character is English. By judging that re-segmentation is necessary when it is not a number, when re-segmenting characters, it is possible to narrow down the characters that may be incorrect and to accurately detect the re-segmentation range Becomes

In the character recognition apparatus according to the second aspect, the character re-slicing determination means recognizes a character rectangle having a character configuration rectangle number of one and a character width longer than a character height. Based on the reliability of the result, the character of interest, character information before and after the character of interest, and the height of the character of interest and the character immediately before the character of interest, it is determined whether re-cutout processing for separating characters can be performed. .

That is, the height of the target character and the character before it are checked, and only when the height of the previous character is the same as that of the character, it is determined that the re-cutting is necessary. It is possible to cut out again even if the processing cannot be solved by the processing of claim 1 using the character type information before and after.

In the character recognition device according to the third aspect, the character re-slicing determination means recognizes a character rectangle having a character configuration rectangle number of one and a character width longer than the character height. Based on the reliability of the result and the ratio of the space (blank) width between the character information before and after the character of interest and the character before and after the character of interest, it is determined whether or not re-cut processing for separating characters can be performed.

That is, the target character and the character spacing before and after the target character are checked, and if the character type of the target character does not match the character type of the character with the smaller blank width of the preceding and following characters,
By judging that re-cutting is necessary, re-cutting can be performed even if the processing cannot be solved by the processing of claim 2.

In the character recognition device according to the fourth aspect, the character re-slicing determination means recognizes a character rectangle having a character configuration rectangle number of one and a character width longer than the character height. Based on the resulting linguistic information (part of speech information)
This is for determining whether or not re-cutout processing for separating characters is executable.

That is, by checking the linguistic information (part-of-speech information) of the recognition result, it is not necessary to re-cut out the correct alphabetic character as a word, so that it is possible to reduce cut-out errors.

In the character recognition device according to the fifth aspect, the character re-cutting determination means determines that re-cutting processing for separating characters again is to be performed, and a character cut-out method different from the first recognition result. When the re-segmentation position is detected, the standard character width is calculated again from the reliability of the recognition result and the character type, the validity of the divided character rectangle width is determined using the standard character width, and whether or not the character re-segmentation process can be performed is determined. Judgment
The apparatus further includes a character re-cutout processing means for processing.

That is, by judging whether or not the re-cut out divided rectangle is appropriate as a character, it is not necessary to apply an abnormal rectangle to the re-recognition processing.

In the character recognition device according to the present invention, the character re-sampling processing means may recognize the re-slicing position if it is determined that the re-slicing position is not appropriate by a character cutting method different from the first recognition result. By using the character type information before and after the result and the blank width before and after the target character, it is determined again whether or not the character re-cutout processing can be executed.

In the character recognition device according to the present invention, when the character re-slicing processing means determines that it is appropriate to divide the target character, the character re-slicing processing means is different from the character extracting method. The re-cutting process is executed by using the character cutting-out method described above.

That is, in claim 6 or claim 7,
Only those determined to be inappropriate by claim 5,
It is possible to improve the efficiency of the process by determining again whether or not the cutout can be executed and executing the cutout again only on the ones that are determined to be appropriate.

In the character recognition device according to the present invention, when a likely result is uniquely selected from a plurality of recognition results having different character cutout positions, a standard is re-determined based on the reliability of the recognition result and the character type. The apparatus further comprises a result evaluation selecting means for determining a character width, evaluating the character using the character type of the result of the re-recognition and the standard character width, excluding inappropriate candidates from a plurality of recognition result candidates, and outputting the result. .

That is, by recognizing a character having an inappropriate character width and not outputting it as a fixed character, only the character suitable for the character is re-cut out, so that the processing efficiency can be improved.

Further, in the character recognition device according to the ninth aspect, the standard character width is obtained from a character having a high degree of reliability of a recognition result in an arbitrary processing target line and a character type of a kanji. .

In other words, in claim 5, 6 or 8, the recognition accuracy is improved by determining the standard character width from a character having a high degree of reliability of the recognition result in an arbitrary processing target line and a character type of Chinese character. It becomes possible.

According to a tenth aspect of the present invention, in the storage medium, a computer is provided with a character re-cutout determining means of the character recognition device according to any one of the first to ninth aspects.
It stores a program for causing it to function as character re-cutout processing means and result evaluation selection means.

That is, by reading the program from the storage medium according to the tenth aspect, character recognition processing can be executed on a computer.

[0030]

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram showing a computer-readable storage medium storing a program for causing a computer to function as a character recognition apparatus according to the present invention.

[Structure of Embodiment] FIG. 1 is a block diagram showing the structure of a character recognition device according to an embodiment. In the figure, the character recognition apparatus includes an image input unit 101 for inputting an image image to be recognized, a line cutout unit 102 for extracting the image image on a line basis, and a line cutout unit 10.
The character extraction unit 103 extracts the character information of the line portion extracted in step 2, the character recognition processing unit 104 that recognizes the character information extracted by the character extraction unit 103, and the character recognition unit 104 On the other hand, a post-processing unit (language processing unit) 105 that performs post-processing including linguistic processing such as morphological analysis, a reliability calculation unit 106 that calculates reliability indicating the certainty of the recognition result, A character re-cutout determining unit 107 as a character re-cutout determining unit for detecting a contact character assumed to be cut out as one character by mistake and determining whether re-cutting of the character can be executed; A character re-cutout processing unit 108 as character re-cutout processing means for dividing a character rectangle for which a character A result evaluation selecting unit 109 as a result evaluation selecting unit for comparing the result with the recognition result after re-cutting and selecting / outputting a likely result, and a fixed character string output unit 110 for fixing character information and outputting it as a character string It is composed of

[Operation of Embodiment] Next, the operation of the character recognition apparatus configured as described above will be described. First, as in the normal character recognition, an image image is input from the image input unit 101, a character recognition process is performed on the image image by a character recognition processing unit 104, and a post-processing unit (language processing unit) 105 performs the same. Post-processing including language processing such as morphological analysis is performed, and a recognition result is obtained. continue,
The reliability calculating unit 106 obtains a reliability indicating the certainty of the recognition result.

The method of obtaining the reliability is generally a method using the distance from the feature amount in the pattern dictionary, the difference from the second candidate character, and the like. I don't care.

Next, an operation which is a feature of this embodiment will be described separately for each of the character re-cutout determination process, the character re-cutout process, and the result evaluation selection process.

Character Re-cutout Determination Process This process is executed by the character re-cutout determination unit 107. That is, a process is performed to detect a character (hereinafter, referred to as a contact character) assumed to be erroneously cut out as one character by touching two characters, and to determine whether or not to re-cut out the character.

Here, what is determined as re-cutout is
The character rectangle is divided, a plurality of character rectangles are generated, and the recognition process is executed again as follows.

First, it is assumed that a character constituting a character has the number of rectangles of 1 and a character width greater than the character height as a contact character, and one of the following three cases is satisfied. In this case, it is determined that the character rectangle needs to be cut out again.

(Case 1) The reliability of the target character is equal to or less than the threshold value, one of the characters before and after the target character is an alphanumeric character, and the target character is not an alphanumeric character. For example, the case of FIG.

(Case 2) When the reliability of the target character is equal to or less than the threshold value, any one of the characters before and after the target character is alphanumeric, and the target character is alphanumeric, the part of speech information of the target character is checked.

The part-of-speech information is provided with information such as a noun, a verb, and an unregistered word to the recognized character by language processing during post-processing. If this information is an unregistered word as shown in FIG. 2B, it is assumed that the character of interest may be erroneous, and it is determined that re-cutout is necessary.

(Case 3) If the part of speech before and after is not an unregistered word, the height of the target pixel and the preceding character are checked. Only when the alphanumeric character of the preceding character and the character height are the same, it is determined that re-cutting is necessary.

This corresponds to, for example, "15 m Ethernet
This is to prevent a case where “m” or the like representing the unit of “et” is erroneously cut out again.

Here, with reference to the flowchart shown in FIG. 3, the character re-sampling processing operation by the character re-slicing determination unit 107 will be described. This processing operation is executed for the recognition result of the character configuration rectangle number of 1 and the character height <character width.

In FIG. 3, first, character reliability <TH
It is determined whether or not cert (reliability threshold: constant) (S301). Here, the reliability of the character <THcer
If it is determined that it is t, it is further determined whether or not the characters before and after the target character are alphanumeric characters (S302).

If it is determined in step S302 that the character before and after the target character is alphanumeric, it is further determined whether or not the target character is alphanumeric (S303). Here, when it is determined that the character of interest is not an alphanumeric character, the re-cutout set is turned ON (S304), and the result is given to the character re-cutout processing unit 108.

If it is determined in step S301 that the reliability of the character is not smaller than THcert, or if it is determined in step S302 that the character before and after the target character is not an alphanumeric character, the re-cutout set is turned off ( S 305), and gives the result to the character re-cutout processing unit 108.

On the other hand, if it is determined in step S303 that the character of interest is an alphanumeric character, it is further determined whether or not the character is an unregistered word (S306). If it is determined that the word is not an unregistered word, it is further determined whether or not the character height is the same as the previous character (S307).

If it is determined in step S306 that the word is unregistered, or if it is determined in step S307
If it is determined that the character height is the same as the previous character, the re-cutout set is turned ON (S304).
This result is provided to the character re-cutout processing unit 108. If it is determined in step S307 that the character height is not the same as the previous character, the re-slicing set is turned off (S305), and the result is provided to the character re-slicing processing unit 108.

Here, the character re-sampling process for dividing the character rectangle into the character rectangles determined to require the re-segmentation process as described above is executed as follows.

FIG. 4 is a flowchart showing an example of the first character re-cutout process according to the embodiment. In the figure, first, the character width is determined in the character width direction (S40).
1), the standard character width is reset (S402). Further, a character division width (hereinafter, referred to as SW1 and SW2) is set by another character extraction method different from the first character extraction method. This character extraction method may be any means. As a character cutting method, for example, a general method of measuring a projection histogram of black pixels in the character width direction and determining a division point may be used.

Next, it is checked whether the divided rectangle reset as described above is appropriate as a character rectangle by using the standard character width reset in step S402. That is,
It is determined whether or not the divided character width ≧ (standard character width / 2) (S403). Here, divided character width ≧ (standard character width /
If it is determined to be 2), the character rectangle is divided by the above-described divided character width (S404). On the other hand, if it is determined that the divided character width is not equal to or more than (the standard character width / 2), the flow shifts to the flow of FIG.

In other words, the standard character width used here is
It is calculated based on the reliability and the character type of the first recognition result.
Then, the average of the character widths of the characters whose recognition result has high reliability and the character type is a kanji is set as the standard character width. Since rectangular characters are used to recognize half-width characters such as alphanumeric characters, the calculated standard character width / 2 value is compared with SW1 and SW2. If SW1 and SW2 are both large character widths, the obtained division width is obtained. Is adopted as it is.

FIG. 5 is a flowchart showing an example of the second character re-cutout processing according to the embodiment. In step S403 in FIG. 4, the divided character width ≧ (standard character width / 2).
If it is determined that the character width is not the same, the divided character width is reset in order to perform the third character segmentation.

In the figure, first, the character type and blank width of the characters before and after the recognition result are checked. Specifically, it is determined whether the preceding character is an alphanumeric character and Lgap <Rgap (S501). Note that Lgap represents the character interval between the target character and the preceding character, and Rgap represents the character interval between the target character and the subsequent character.

If it is determined in step S501 that the preceding character is an alphanumeric character and Lgap <Rgap, the character width is set by another method to divide the character rectangle (S502).

On the other hand, if it is determined in step S501 that the preceding character is alphanumeric and that Lgap <Rgap is not satisfied, the succeeding character is alphanumeric and Lgap> Rg
It is determined whether it is ap (S503). If it is determined that the characters are alphanumeric and Lgap> Rgap,
The divided character width is set by another method, and the character rectangle is divided (S502). In addition, alphanumeric characters and Lgap> Rga
If it is not p, re-cutout is not performed (S5
04), this flow ends.

That is, here, it is intended to extract a character with the same size as the character type before or after the narrower blank width.

For example, in the case shown in FIG.
The space width with the character “1” before the target character “plate” is narrower than the space width with the subsequent character “East”. From this, it is assumed that the character of interest is more suitable for the character width of the previous character “1”,
Divides characters according to the previous character width. At this time, the character is divided by a method different from the character extraction performed in FIG.

Next, the operation of the result evaluation selection unit 109 will be described. In this case, a likely result is selected by comparing the recognition result before re-cutting the character rectangle with the recognition result after re-cutting. At that time, the recognition result with inappropriate character width is excluded,
Do not output as fixed characters.

That is, when a likely result is uniquely selected from a plurality of recognition results having different character cutout positions, the standard character width is calculated again from the reliability and the character type of the recognition result, and the character type of the re-recognized result is obtained. Then, an inappropriate character candidate is excluded from the plurality of recognition result candidates using the obtained standard character width.

The standard character width is set again based on the above-described character re-cut processing. That is, it is obtained from a character having a high degree of reliability of a recognition result in an arbitrary processing target line and a character type of a Chinese character.

Further, when the character width of the recognition result is smaller than the standard character width and the character type of the recognition result is a kanji, it is highly probable that the character extraction method is erroneous. And not.

For example, in a case where the recognition result character before re-cutout is “read”, the processing is executed in the following procedure. → Character rectangle division → Recognition result characters “word” and “sell” after re-slicing → Standard character width measurement / check → standard character width> character width of “word” Standard character width> character width of “sell” → re-slicing "Word" and "Sell" are not selected as later recognition result characters.

[0064]

As described above, according to the character recognition device of the present invention (claim 1), the number of rectangles constituting a character is one, and the character width is the height of the character. Assuming that a character that exceeds the character is a contact character candidate that is assumed to be cut out as one character by contacting two characters, the reliability of the target character is equal to or less than the threshold value, and one of the characters before and after the target character is alphanumeric, If the target character is not alphanumeric and it is determined that re-segmentation is necessary, when re-segmenting characters, narrow down the characters that may be incorrect and accurately detect the re-segmentation range. Is possible, and the re-cutout efficiency is improved.

Further, according to the character recognition device of the present invention (claim 2), the height of the target character and the character before it are checked, and only when the height of the previous character is the same as that of the character, Since it is determined that clipping is necessary, re-clipping can be performed even in a case where the reliability cannot be solved by using the reliability of the target pixel and the character type information before and after.

Further, according to the character recognition device of the present invention (claim 3), the character of interest and the character spacing before and after the character of interest are checked. By determining that clipping is necessary, it is possible to perform re-clipping even when the processing cannot be solved by the processing of claim 2.

Further, according to the character recognition device of the present invention (claim 4), by checking the linguistic information (part-of-speech information) of the recognition result, it is not necessary to re-cut out a correct alphabetic character as a word. In addition, cutting errors can be reduced, and efficient re-cutting can be realized.

According to the character recognition device of the present invention (claim 5), it is necessary to determine whether or not the re-cut segmented rectangle is appropriate as a character so that the abnormal rectangle is subjected to the re-recognition processing. Is eliminated, and cutout errors can be reduced.

Further, according to the character recognition device of the present invention (claims 6 and 7), it is determined again whether or not cutout can be executed only for those determined to be inappropriate by claim 5,
Since re-cutting is executed only for those determined to be appropriate, processing efficiency can be improved.

Further, according to the character recognition device of the present invention (claim 8), a recognition result having an inappropriate character width is excluded, and the character is not output as a fixed character. Since the execution is performed, the efficiency of the processing can be improved.

According to the character recognition device of the present invention (claim 9), in claim 5, 6 or 8, the standard character width is set to a value that is high in the reliability of the recognition result in an arbitrary processing target line, and Since the character type is obtained from characters whose characters are kanji, the recognition accuracy can be improved.

The storage medium according to the present invention (claim 1)
According to 0), a program for causing a computer to function as a character re-cutout determination unit, a character re-cutout processing unit, and a result evaluation selection unit of the character recognition device according to any one of claims 1 to 9. Is stored, the character recognition processing can be executed on the computer, and the re-segmentation efficiency can be improved.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a character recognition device according to an embodiment.

FIG. 2 is an explanatory diagram showing an example of a contact character and an example of cutting out the character according to the embodiment.

FIG. 3 is a flowchart illustrating a processing operation of a character cutout determination unit according to the embodiment.

FIG. 4 is a flowchart showing an example of a first character re-cutout process according to the embodiment;

FIG. 5 is a flowchart illustrating an example of a second character re-cutout process according to the embodiment;

[Explanation of symbols]

 Reference Signs List 101 Image input unit 102 Line cutout unit 103 Character cutout unit 104 Character recognition processing unit 105 Post-processing unit (language processing unit) 106 Reliability calculation unit 107 Character recutout determination unit 108 Character recutout processing unit 109 Result evaluation selection unit 110 Confirmation String output section

Claims (10)

[Claims] 1. A character string and a character rectangle are cut out from input character image information to recognize the character rectangle, a reliability indicating a certainty of the recognition result is obtained, and the character rectangle is determined in accordance with the reliability. In the character recognition device for re-cutting out and recognizing and comparing and evaluating the first recognition result and the re-cutting-out result and selecting and outputting a likely result, the number of rectangles constituting the character is 1 and the character width is Re-cutout determining means for determining whether or not re-cutout processing for separating characters can be executed based on the reliability of the recognition result and the target character and character information before and after the target character for a character rectangle longer than the character height A character recognition device comprising: 2. The character re-slicing determining means determines the reliability of the recognition result, the character of interest, and the character before and after the character of interest for a character rectangle whose number of constituent rectangles is one and whose character width is longer than the character height. 2. The character recognition device according to claim 1, wherein whether or not re-segmentation processing for separating characters is to be performed is determined based on the character information and the height of the target character and the character immediately before the target character. 3. The character re-slicing determining means determines the reliability of the recognition result and the character information before and after the character of interest for a character rectangle whose number of constituent rectangles is 1 and whose character width is longer than the character height. 2. The character recognition apparatus according to claim 1, wherein whether or not re-segmentation processing for separating characters is to be executed is determined based on a ratio of an interval (blank) width between the character before and after the character of interest. 4. The character re-slicing determining means determines a character rectangle based on linguistic information (part-of-speech information) of a recognition result with respect to a character rectangle having a character configuration rectangle number of 1 and a character width longer than the character height. 2. The character recognition device according to claim 1, wherein it is determined whether or not re-cutout processing for separating the character string can be executed. 5. When the character re-sampling determining means determines that re-segmentation processing for separating characters again is to be performed and a re-segmentation position is detected by a character segmentation method different from the first recognition result, the reliability of the recognition result is determined. And a character re-cut processing means for determining again the standard character width from the character type and the standard character width, determining the validity of the divided character rectangle width using the standard character width, and judging and processing whether or not the character re-cut processing can be executed. The character recognition device according to claim 1, wherein: 6. The character re-cutout processing means, when it is determined that the re-cutout position is not appropriate by a character cutout method different from the first recognition result, character type information before and after the recognition result and a blank before and after the target character. 6. The character recognition device according to claim 5, wherein whether or not the character re-cutout processing can be executed is determined again using the width. 7. The character re-segmentation processing means, when judging that it is appropriate to divide a character of interest, executes re-segmentation processing using a third character segmentation method different from the character segmentation method. 7. The method according to claim 6, wherein
The character recognition device according to 1. 8. When a likely result is uniquely selected from a plurality of recognition results having different character cutout positions, a standard character width is obtained again from the reliability of the recognition result and the character type, and the character type of the re-recognized result and the character type are determined. 8. The apparatus according to claim 1, further comprising a result evaluation selecting unit that evaluates using the standard character width and outputs an inappropriate candidate from a plurality of recognition result candidates. Character recognition device according to the description. 9. The character according to claim 5, wherein the standard character width is determined from a character having a high degree of reliability of a recognition result in an arbitrary processing target line and a character type of a Chinese character. Recognition device. 10. A program for causing a computer to function as character re-cutout determination means, character re-cutout processing means and result evaluation selection means of the character recognition device according to any one of claims 1 to 9. A computer-readable storage medium having stored therein.