JP4334068B2 - Keyword extraction method and apparatus for image document - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to an image document keyword extraction method, and more particularly, to a keyword extraction method for extracting a keyword from an image document recognized by OCR character recognition.
[0002]
[Prior art]
Due to the recent spread of PCs, infrastructure development centered on the Internet, etc., and increased environmental awareness such as forest resource protection, information transmission and storage methods were mainly digital information instead of conventional paper. It is changing into things. Good examples are information exchange by e-mail, browsing of information on web, and publication. However, the vast amount of information based on conventional paper remains an intellectual asset for both individuals and businesses, and cannot be discarded. Therefore, information on paper must be converted into digital information so that it can be used somehow.
[0003]
The conversion into the digital information is generally performed as follows. First, a paper document (meaning that the document information medium is paper) is read from the scanner and converted into an image document. If it is just digital information, it can be left as it is. However, in order to be able to search the contents of the document, it is necessary to convert the image document into code information using OCR (Optical Character Recognition) (to the image document). If there is text information). When a large number of documents are managed and a large number of unspecified persons use the documents, it becomes easy to grasp the outline when searching and browsing the documents by using information such as summary sentences or keywords.
[0004]
As described above, it is possible to convert a paper document into effective digital information by “scan + OCR + keyword / summary sentence extraction”. However, OCR cannot guarantee a 100% recognition rate for English sentences as well as Japanese sentences with complex shapes such as kanji characters, so the code information converted by OCR includes erroneous codes. It is always done. If the quality of the scanned image is not good due to the state of the paper document or the scan surface of the scanner, the text converted by OCR contains a high percentage of errors. When keyword extraction is performed from text that includes such an error, there is a high possibility that the extracted keyword includes an erroneous character.
[0005]
Although automatic OCR performs OCR without human intervention, if a keyword is extracted by correcting OCR misrecognized characters sequentially through manual operation, the erroneous recognition characters are not included in the keyword. However, considering the time from scanning a paper document to keyword extraction, manual OCR is not practical because it takes too much processing time and manpower. Automatic OCR outputs an OCR result file as a conversion result. The OCR result file is a binary file including the certainty factor and character position information for each character.
[0006]
As described above, generally, when keywords are extracted from text generated by subjecting an image document to OCR, a character string composed of misrecognized characters may be included in the keyword list. This problem seems to be solved by taking a misrecognition countermeasure that excludes characters with low confidence in the recognition result of the OCR from the text. However, the following problem actually occurs.
[0007]
(1) If a character string that has been processed by eliminating some misrecognized characters becomes a keyword, and that character string is a character string that cannot be confirmed in the original image document, it is a problem for the user.
(2) To process the character string that can be confirmed by the user, at least the text to be passed to the keyword extraction unit needs to be divided into words. That is, it is necessary to check the word length and determine whether there is no problem even if up to how many characters are misrecognized. For that purpose, the automatic OCR needs to perform morphological analysis after converting the image into text, and the processing system becomes heavy.
[0008]
The present invention has been made in view of the above circumstances, and ensures that even if there is an error in the text converted by OCR, the extracted keyword has only an allowable error (for example, with 6 characters). An object of the present invention is to provide an image document keyword extraction method and apparatus which can easily understand what this keyword means even if there is an error in one character of the constructed keywords. Is.
[0009]
[Means for Solving the Problems]
According to the first aspect of the present invention, an OCR in which a candidate for character information including a character code and certainty information of the character code is generated for each character by OCR character recognition of the image document by the plain text and certainty factor file generating means. The result file is input, and the character code and the certainty information included in the first candidate character information are extracted for each character from the character information candidates, and the plain text based on the character code and the certainty information are used. A step of generating a certainty factor file, a step of generating a keyword list by performing morphological analysis and keyword extraction of the obtained plain text by the keyword extracting means, and a step of generating the keyword list by the keyword verifying means Whether the keyword is misidentified based on a preset threshold By character is determined, it is obtained by, comprising the steps of: a keyword performs processing to remove from the keyword list containing more false characters than the number calculated-out based on a predetermined condition.
[0012]
The invention of claim 2 refers to the entry of the certainty factor list corresponding to the keyword included in the keyword list in the invention of claim 1, and specifies position information of the keyword in the image document from the referenced entry. And a step of highlighting the keyword in the image document based on the specified position information.
The invention according to claim 3 inputs an OCR result file in which character information candidates including character codes and certainty information of the character codes are generated for each character by OCR character recognition of the image document. Plain text for extracting the character code and the certainty factor information included in the character information of the first candidate for each character from the candidates, and generating a plain text based on the character code and a certainty factor file based on the certainty factor information Certainty factor file generation means, keyword extraction means for generating a keyword list by performing morphological analysis and keyword extraction of the obtained plain text, and keywords set in the keyword list generated by the keyword extraction means are preset. Based on the specified threshold value, it is judged one by one whether the character is mistaken, and based on the predetermined condition. Is obtained by comprising: the keyword verification means for a keyword containing a calculated often misidentified characters than the number can perform a process of removing from the keyword list, the.
The invention of claim 4 refers to the entry of the certainty factor list corresponding to the keyword included in the keyword list in the invention of claim 3 , and specifies position information of the keyword in the image document from the referenced entry. And a means for highlighting the keyword in the image document based on the specified position information.
[0013]
DETAILED DESCRIPTION OF THE INVENTION
Before describing the operation of the present invention, an outline of the structure of the OCR result file and the keyword list will be described. FIG. 1 is a diagram illustrating an example of a TAG structure of an OCR result file.
[0014]
There are the following three typical TAGs.
(1) Information such as image information (resolution, size, number of areas, etc.) for one page is stored in the area pointed to by the page information tag start offset.
(2) Information such as the position of the area is stored in the area pointed to by the area information tag start offset. Since the area may be nested, it has a slightly different structure from that of the page information tag. However, this is not directly related to the present invention and will not be described.
(3) Character information tag Information such as a character code as a recognition result, a certainty of the recognition result, and a recognition coordinate position (position of a rectangle surrounding the character: pixel value) is stored as character information in one area. .
[0015]
Assuming that n characters exist in the area, the character information tag and the character information area are structured as shown in FIG. 2, and the recognition result (Ci) of the i-th character at this time is shown in FIG. The data structure is as follows. Regarding character information, since a plurality of recognition results are obtained for one character, there are a plurality of candidates for one character.
[0016]
(Description of operation)
FIG. 4 is a diagram for explaining a schematic flow of keyword extraction processing to which the present invention is applied. The paper document 1 is scanned 2 by the scanner, converted into an image document 3, processed by automatic OCR 4, and an OCR result file 5 is output. The obtained OCR result file 5 is processed by the keyword extraction unit 6 and a keyword list 7 is output. FIG. 5 is a diagram showing in more detail the keyword extraction unit 6 that performs the keyword extraction of the present invention shown in FIG. Below, the process in the keyword extraction part 6 is demonstrated in detail.
[0017]
[1] In the plain text / confidence file generation unit 6a as the pre-processing pre-processing, the character code and the certainty factor of the first candidate are extracted from the OCR result file 5 obtained by the automatic OCR 4, and the plain text 6c and the certainty factor file are obtained. 6b is generated. The plain text 6c is passed to the keyword extraction unit 6e. The certainty factor file 6b has a structure as shown in FIG. 6, and holds the certainty factor of only the first candidate for each character.
[0018]
If the input image document is one sheet, it is sufficient to refer to the OCR result file directly and refer to the certainty factor. Otherwise, the keyword extraction ends for the OCR result file of the input image constituting the document. Since it is necessary to keep everything until, create a confidence file separately. The certainty factor file 6b is created for each page, and only one plain text 6c is created for each document. Each entry in the certainty factor file 6b and the characters in the plain text 6c need to be completely synchronized.
[0019]
[2] Keyword extraction unit The keyword extraction unit 6e performs morphological analysis and syntactic analysis of texts, and mainly shows noun phrases, appearance frequency, similarity check, idiomatic phrase check, modification in sentences, dependency, etc. Keywords are extracted from the information, and a keyword list is generated. Here, the keyword extraction unit 6e accepts only plain text and does not accept non-text files such as a certainty factor file. An outline of the structure of the generated keyword list is shown in FIG. The keyword list output from the keyword extraction unit 6e is the keyword list 6g including misidentification.
[0020]
[3] As post-processing, the keyword verification unit 6f inspects one character at a time whether or not there is a misidentified character from the upper keywords in the keyword list. Whether a character is mistaken is determined by a threshold value. The threshold is determined dynamically for each image. Since the keyword extraction unit 6e also discloses the keyword position to the application as keyword information, the keyword start position in the plain text that is the input of the keyword extraction unit 6e can be known.
[0021]
Since the OCR result file 5 and the plain text 6c are synchronized with the recognized character, the certainty factor of the character can be understood by referring to the OCR result file 5. As shown in FIG. 8, keywords including characters with low certainty are removed from the keyword list, and the ranking of the next keyword is increased. If this operation is repeated until the desired number of keywords is reached, a keyword list 7 that does not include misidentified characters can be obtained.
[0022]
Judgment formulas such as the following are used to determine whether or not to leave the keyword list.
Number of misidentified characters in keyword> number of characters constituting keyword × (α / 100)
Α is a% value given by the user. For example, if a keyword including 10% or more of misidentified characters is not required, 10 is designated as α. In reality, if the keyword is a four-character keyword including about 1 misidentified character, the user can discriminate the character, so the misidentified character is replaced with a symbol such as “?” And presented to the user (of the 2 characters). If one character is mistaken, it cannot be replaced with “?”. As described above, readable keywords can be acquired from the image document in the form of text codes.
[0023]
[4] Highlighting keywords in an image document If the character position information (image position information) in a certainty factor file is used, it is possible to display the extracted keywords on the image document. Since each entry in the certainty factor file and each entry in the plain text have a one-to-one correspondence, if the keyword list indicates the position of the keyword in the plain text, the keyword position in the certainty factor file is also known. If the position on the certainty factor file is known, the position of the first character of the keyword in the page of the image document can be known. Similarly, the position of each character constituting the keyword is also known. Since each character position is expressed by a rectangle, when emphasizing a keyword, an OR mask may be applied with an RGB value representing an appropriate color for the rectangular region surrounding the character. It is also possible to change only the color of the character by a bit operation.
[0024]
[5] Image quality of an image document When the image quality of an image document is poor, a large number of misidentified characters are included in the OCR result file. In such a situation, even if keywords are extracted by the above methods [1] to [3], they may belong to a group of keywords belonging to the lower level of the keyword list or there may be no keywords in the keyword list. The extraction process becomes meaningless. In such a case, a statistical value (standard deviation or the like) 6d of the certainty factor for each character in the certainty factor file is calculated for each certainty factor file, and the magnitude relationship between the statistical value and the “threshold value” is compared. It is determined whether or not to perform keyword extraction.
[0025]
This statistic also determines a “threshold” that is used to determine whether a character in the OCR result file is misidentified.
Character misidentification threshold = Func (statistical value of confidence file)
A specific definition formula of Func (statistical value of confidence file) is omitted. This equation indicates that the statistical value of the certainty factor file affects the threshold for character misidentification. The “threshold value” is dynamically determined for each image. Since the state of the image is different for each image, it is a reasonable method to quantify the image quality from the OCR result.
[0026]
【The invention's effect】
According to the present invention, even if there is an error in the text converted by OCR, it is guaranteed that the extracted keyword has only an allowable error (for example, one of the six-character keywords is erroneous). Even if there is, it is possible to easily understand what this keyword means), and the convenience of the user can be enhanced without increasing the processing system.
[Brief description of the drawings]
FIG. 1 is a diagram illustrating an example of a TAG structure of an OCR result file.
FIG. 2 is a diagram illustrating an example of a structure of a character information tag and a character information area in an OCR result file.
FIG. 3 is a diagram illustrating an example of a data structure of an i-th character illustrated in FIG.
FIG. 4 is a flowchart for explaining the outline of the processing operation of the keyword extracting method according to the present invention.
FIG. 5 is a diagram for explaining the keyword extraction unit shown in FIG. 4 in more detail.
FIG. 6 is a diagram illustrating an example of the structure of a certainty factor file.
FIG. 7 is a diagram illustrating an example of the structure of a keyword list.
FIG. 8 is a diagram for explaining processing for removing a keyword from a keyword list.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 ... Paper document, 2 ... Scan, 3 ... Image document, 4 ... Automatic OCR, 5 ... OCR result file, 6 ... Keyword extraction part, 6a ... Plain text and certainty factor file generation part (preprocessing), 6b ... Certainty factor File, 6c ... Plain text, 6d ... Statistical value of certainty factor, 6e ... Keyword extraction unit, 6f ... Keyword verification unit (post-processing), 6g ... Keyword list including misidentification, 7 ... Keyword list.

Claims (4)

A plain text / confidence file generating means inputs an OCR result file in which character information candidates including character codes and certainty information of the character codes are generated for each character by OCR character recognition of the image document, The character code and the certainty factor information included in the first candidate character information are extracted for each character from the character information candidates, and a plain text based on the character code and a certainty factor file based on the certainty factor information are generated. A step of generating a keyword list by performing morphological analysis and keyword extraction of the obtained plain text by a keyword extraction means ;
The keyword verification means, with respect to keywords generated the keyword list, it is determined whether the misidentified a character one by one on the basis of a preset threshold, than the number calculated-out based on a predetermined condition And a step of removing a keyword including a large number of misidentified characters from the keyword list.   The keyword extraction method for an image document according to claim 1, wherein an entry in the certainty factor list corresponding to a keyword included in the keyword list is referenced, and position information of the keyword in the image document is specified from the referenced entry. A method for extracting a keyword of an image document, comprising the step of highlighting the keyword in the image document based on the specified position information. An OCR result file in which a character information candidate including a character code and certainty information of the character code is generated for each character by OCR character recognition of the image document is input. A plain text and certainty factor file generating means for extracting the character code and the certainty factor information included in the candidate character information for each character, and generating the plain text by the character code and the certainty factor file by the certainty factor information, Keyword extraction means for generating a keyword list by performing morphological analysis and keyword extraction of the obtained plain text,
With respect to the keyword list keyword generated by the keyword extraction unit, determines whether the misidentified a character one by one on the basis of a preset threshold, than the number calculated-out based on a predetermined condition And a keyword verification unit that performs processing for removing a keyword including a large number of misidentified characters from the keyword list. 4. The keyword extracting apparatus for an image document according to claim 3 , wherein an entry in the certainty factor list corresponding to a keyword included in the keyword list is referenced, and position information of the keyword in the image document is specified from the referenced entry. An image document keyword extraction apparatus comprising means for highlighting the keyword in the image document based on the specified position information.

Images