JP2977247B2 - Inter-character space processing method - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for processing a space between characters in a character recognition device, and more particularly to a method for processing a space between characters applied to a kanji OCR or the like for processing Japanese sentences. About.

[Conventional technology]

Generally, in a character recognition device, the detection of an inter-character space (space character) is performed based on character cut-out information. For example, it is determined whether or not the space between characters is a space by comparing the width of the space between characters with the print pitch. Then, the detected space is output as a recognition result as it is.

[Problems to be solved by the invention]

In the case of Japanese sentences, half-width numbers, alphabets, and symbols are used in the text, as well as full-width kanji and hiragana characters, and the space includes both full-width and half-width spaces.
In order to obtain a recognized output sentence that is faithful to the Japanese sentence, it is necessary to recognize and output a space and other characters by distinguishing half-width and full-width characters.

However, when trying to detect a half-width space in a Japanese sentence by the conventional method using the space between characters, a space that did not exist in the input sentence was generated during the recognition output, and the recognition output sentence was input. Sometimes it was different from the text. The reason is that in the case of Japanese sentences, there are not many characters such as double-byte kanji, hiragana, and symbols separated left and right, and characters and symbols printed unevenly on one side. And blank portions on both sides are easily erroneously detected as spaces.

For example, the central space of the hiragana "i" is detected as a half-width space, or in the sentence "... is next ...", the punctuation mark "." Is printed eccentrically to the left. The phenomenon of being detected as a space is likely to occur. A similar phenomenon is caused by parentheses (“”, “”
() [] ｛｝ <>), Punctuation marks (,.,. ""
It is easy to occur before and after '').

SUMMARY OF THE INVENTION It is an object of the present invention to provide an inter-character space processing method for eliminating an extra space in a kanji OCR and obtaining a recognized output sentence faithful to an input sentence.

Also, depending on the application that uses character recognition, when the presence of a half-width code is not preferable in data processing (such as when using the recognition result as input to a database) or when less space is preferable (such as information compression of the recognition result). There is also.

It is another object of the present invention to be able to output a recognition result in a format suitable for such a case.

[Means for solving the problem]

The present invention relates to a space between characters detected based on character cut-out information, the character type of one or both of characters before and after the detected space, and the type of the detected space (half-width space, The main feature is to perform processing of validity (the detected space remains as it is), invalidity (no space), or conversion (half-width space to full-width space, or vice versa) based on the full-width space.

Further, the present invention is characterized in that a plurality of rules for validating, invalidating or converting the detected space are provided, and one or more rules to be applied are selected from the rules. Further, the present invention is characterized by including a rule for invalidating a half-width space or a full-width space unconditionally, or for converting a half-width space to a full-width space unconditionally.

(Operation)

As described above, when a space is detected by a space detection method based on character cut-out information, for example, a method of comparing a space width between individual characters and a full-width character standard width, an unnecessary half-width space is detected or a half-width space is detected. May be detected as whole space. However, such an unnecessary half-width space or an error between a half-width space and a full-width space can be predicted in advance from the relationship between the characters before and after the half-width space.

According to the present invention, the recognition result of one or both characters before and after the space detected based on the character cutout information,
Furthermore, by examining the relationship between the detected space and the recognition result, it is possible to determine whether the space is valid or whether the half-width space and the full-width space are incorrectly taken, and correct the space. Thus, it is possible to correct most of the space detection errors occurring in Japanese sentences by the space detection based on the character extraction information.

Here, the following are conceivable as specific examples of rules for validating, invalidating, or converting the detected space.

Rule A: The single-byte space is valid only when the recognition result of both characters before and after the single-byte space is alphanumeric.

Rule B: The single-byte space is valid only when the recognition result before and / or after the single-byte space is an alphanumeric character.

Rule C: When the recognition result of the character before the one-byte space is a punctuation mark (,., .., etc.) and there is any character after that, the one-byte space is invalidated.

Rule D: When the recognition result of the character before the one-byte or two-byte space is a punctuation mark, and there is some character after it, the one-byte or two-byte space is invalidated.

Rule E: The recognition result of the character after the single-byte space starts and the parenthesis (｛(“[[, etc.], if there is any character before, invalidates the single-byte space.

Rule F: A recognition result of a character after a double-byte or single-byte space starts and a parenthesis (｛(“[[, etc.) is used to invalidate the double-byte or single-byte space when there is any character before it.

Rule G: When the recognition result of the character before the half-width space ends with parentheses ("") {), etc., and there is any character after that, the half-width space is invalidated.

Rule H: When the recognition result of the character before the double-byte or single-byte space ends with parentheses ("") {), etc., and when there is any character after it, the double-byte or single-byte space is invalidated.

Further, there is a case where it is desired to completely remove a space or a half-width space from a recognition result output and obtain a recognition result output convenient for an application using the recognition result output. The following rules can be considered as specific rules applied in such a case.

Rule I: One-byte space is invalid regardless of the recognition result of characters before and after it.

Rule J: A double-byte or single-byte space is invalid regardless of the recognition result of characters before and after.

Rule K: A half-width space is converted to a full-width space regardless of the recognition result of the characters before and after it.

〔Example〕

FIG. 1 is a block diagram of a character recognition device according to one embodiment of the present invention. In this character recognition device, the image input unit
10 reads the image of the original by a scanner or the like, and
The value image data is input and stored in the image memory 11. line·
The character cutout unit 12 cuts out a character line and a character image from the input image in the image memory 11, stores the character image data in the character image memory 13, and outputs a character cutout position, a character width, a width of a space between characters, and a line. The cutout information such as the height of the character and the standard width of the character is stored in the cutout information memory 14.

The space detecting unit 15 detects a half-width space or a full-width space between characters based on the cut-out information obtained in the cut-out information memory 14, and outputs the detection result information to the cut-out information memory 14.
Is stored corresponding to the inter-character space.

The character recognition unit 16 reads character image data from the character image memory 13, performs normalization processing, and then extracts features,
The extracted features are compared with the features of the dictionary read from the dictionary memory 17, and a recognition result candidate having a small feature distance is obtained and stored in the recognition result memory 18. Further, referring to the space detection result information in the cut-out information memory 14, information on half-width or full-width spaces between the detected characters is also stored in the recognition result memory 18 in accordance with the character arrangement. Therefore, the recognition result memory 18
In the table, there is obtained recognition result data in which character recognition result candidates and space codes are arranged in line with the character arrangement on the document.

The space processing unit 19 applies the space processing rule to the full-width or half-width space obtained in the recognition result memory 18, thereby invalidating unnecessary space or converting inappropriate half-width space to full-width space, or Processing such as unconditional deletion (invalid) of spaces and unconditional conversion of half-width spaces to full-width spaces is performed, and the recognition result memory 18
Rewrite the contents of The rule storage unit 20 stores rules applicable to space processing. The user of the apparatus selects one or a combination of two or more rules to be applied to the rule processing from a plurality of rules (for example, the above rules A to K) stored in the rule storage unit 20. Can be. This selection can be made through a console not shown.

The result output unit 21 outputs the recognition result data after the space processing in the recognition result memory 21 to an output device such as a disk or a printer.

Next, the horizontal character string “This is the“ new
AIBCH @ WP "(for convenience, the half-width space is indicated by a triangle and the full-width space is indicated by a square. The same applies to the following description), and the contents of space detection and space processing will be described.

The space detection unit 15 refers to the cutout information memory 14,
By comparing the space width between characters with the full width character standard width,
Detects single-byte and double-byte spaces. The full-width character standard width is calculated in the line / character cutout unit 12 (or in the space detection unit 15) using the height of the line, the average value of the character width in the character string, and the like.

The specific detection processing procedure is as shown in the flowchart of FIG. 3. In this case, the inter-character space width is divided by the full-width character standard width in order from the first inter-character space, and if the division result is 1.0 or more. The inter-character space is detected as a full-width space, and if the division result is less than 1.0 and 0.5 or more, the inter-character space is detected as a half-width space, and if the division result is less than 0.5, it is determined that the space is not a space. This is repeated until the last blank of the line, and the result is written into the cut-out information memory 14.

The contents of the cutout information memory 14 for the character string in the example are the first.
Assume that it is as shown in the table.

In this case, since the full-width character standard width is 60, a space between characters having a width of 60 or more is determined to be a full-width space, and a space between characters having a width of 30 to 59 is determined to be a half-width space.

Therefore, this character string is "This is □" New AIBCH
The first full-width space and the last half-width space are extra spaces that do not exist in the input character string. Such extra spaces are caused by punctuation and parentheses skewed to the left or right. The detected space, including such extra space, is also inserted into the recognition result character string in the recognition result memory 18 as it is.

Next, the processing of the space processing unit 19 will be described. A general suitable flow of this processing is as shown in the flowchart of FIG. However, it is not always necessary to determine half-width or full-width depending on the rule to be applied. In addition, NG
(Invalid, conversion required, etc.)
It depends on the rules applied.

For example, when the rule B “the half-width space is valid only when both or one of the recognition results before and after the half-width space is an alphanumeric character” is applied, the character before and after the last half-width space in the example character string is “”. Since both "" and "" are not alphanumeric, they are determined to be invalid, and the data of the space in the recognition result memory 18 is rewritten to "none".

Therefore, the character string is “This is □“ New △ AIBCH △ WP ”
Is corrected to a closer result.

In addition, when the rule D “the recognition result of the character before the double-byte or single-byte space is a punctuation mark and there is any character after it, the double-byte or single-byte space is invalid” is applied simultaneously with the rule B. The first double-byte space in the character string is determined to be invalid because the punctuation mark "," precedes it and the parenthesis "[" follows.

Therefore, the character string is "This is 'new AIBCH WP'", and the correct result was obtained.

The other rules described above can be applied alone or in combination. For example, by applying the rules F and H, spaces before and after parentheses in a character string can be deleted. If the application that uses the recognition result output dislikes the presence of a half-width space, the rule I or K can be applied to delete (invalidate) all the half-width spaces or convert them to full-width spaces. Further, by applying the rule J, all double-byte spaces and single-byte spaces in a character string can be deleted.

In this way, by preparing various rules in advance, according to the format of the input sentence and the purpose of outputting the recognition result,
By selecting necessary rules and applying them alone or in combination, it is possible to easily obtain a recognition result output subjected to appropriate space processing.

〔The invention's effect〕

As described above, according to the inventions described in claims (1) and (2), it is possible to delete (invalidate) an extra space detected based on character cut-out information and correct mistaking of a half-width space and a full-width space. It is possible to determine and correct most of space detection errors occurring in Japanese sentences.
According to the invention described in claim (3), by selecting one or a combination of rules to be applied in accordance with the format of the input sentence, the method of using the output of the recognition result, and the like, the output of the recognition result in a desired format is achieved. Can be easily obtained. Further, according to the invention described in claim (4), it is possible to completely eliminate a space or a half-width space from the recognition result output, so that a recognition result output convenient for an application using the recognition result output can be provided. Can be easily obtained.

[Brief description of the drawings]

FIG. 1 is a block diagram of a character recognition device according to one embodiment of the present invention, FIG. 2 is an explanatory diagram of an example of an input character string, FIG. 3 is a flowchart of space detection processing, and FIG. It is. 10 ... image input unit, 11 ... image memory, 12 ... line / character cutout unit, 13 ... character image memory, 14 ... cutout information memory, 15 ... space detection unit, 16 ... character recognition unit, 17: Character dictionary memory, 18: Recognition result memory, 19: Space processing unit, 20: Rule storage unit, 21: Result output unit.

Claims (4)

(57) [Claims] An object of the present invention is to detect a space between characters based on character cut-out information, and to perform valid, invalid, or conversion on the detected space based on one or both character types of characters before and after the space. Characteristic space handling method. 2. A method for detecting a space between characters based on character cut-out information, and detecting whether the detected space is valid, invalid, or invalid based on a type of the space, one or both of characters before and after the space. A method for processing a space between characters, comprising performing conversion. 3. The method according to claim 1, wherein the detected space is effective.
A method for processing a space between characters, comprising a plurality of rules for performing invalidation or conversion, and selecting one or more rules to be applied from the rules. 4. A method according to claim 3, further comprising a rule for invalidating a half-width space or a full-width space or for converting a half-width space to a full-width space unconditionally. Character space processing method characterized by the following.