JP3487523B2 - Document processing device - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a document processing apparatus for detecting a logical structure of a document image.

[0002]

2. Description of the Related Art A document processing apparatus such as an electronic file system can read and input a document image with a scanner and store it in a database. When the document image is stored in the database as described above, even if the character is simply read and input as a pattern, it can be recognized by a person by displaying and outputting it, which is useful.

The document image stored in the database as described above cannot be effectively used unless it can be freely searched. In order to freely search the document image stored in the database, it is necessary to set a search keyword in the document image. However, in order to set a keyword in the document image stored in the database, it is necessary for the operator to read an appropriate term from the document image and perform an input operation using the keyboard.

In order to solve such a problem, Japanese Unexamined Patent Publication No.
In the device disclosed in Japanese Patent No. 159101, the logical structure of the document image read and input by the scanner is detected, and the description content of the document image is recognized according to this logical structure. More specifically, a plurality of structural models are set in advance, and the relationship between the layout structure of the document image and the logical structure is defined in this structural model. A document image read and input by a scanner is divided into a plurality of elements, and a structural model matching the mutual relationship of the divided plurality of elements is detected.
Since the logical structure of the document image is recognized by the detection of the structural model, each of the plurality of elements can be identified as a title, an author name, a body, and the like, and an optimum term for the keyword can be extracted.

[0005]

When the document image is input and saved as described above, it is desirable to recognize the logical structure in order to extract the keyword.

However, in the apparatus disclosed in Japanese Patent Laid-Open No. 5-159101, the position of an element on a page and the relative positions of a plurality of elements are assumed, and the logical structure is recognized based on this. Therefore, the logical structure of a document image whose format is fixed, such as academic papers, can be recognized well,
It is difficult to recognize a logical structure from a document image having various layouts such as an office notification document.

[0007] Specifically, in the case of office notification documents,
Bibliographic information such as a title, a date, and a sender is written at a specific position, but the date may be omitted, for example. Since the format of the text varies widely, it is difficult to match this with the structural model. The image quality may be deteriorated due to repeated facsimile transmission or copying, and in this case, character recognition is difficult, so that the logical structure cannot be recognized well. In some cases, handwritten memos and stamps are added, in which case the accuracy of element division and character recognition is reduced.

When using the document image stored in the database, the document image is read out from the database, displayed on the display, and the described content is confirmed, and only the required document image is printed out by the printer. There are many. In such a case, since the document image read and input by the scanner and stored in the database is image data, it is difficult to confirm the description content if the display has low resolution.

In such a case, if the title, author name, etc. of the document image are coded, they can be displayed well even on a low-resolution display, so that the description content of the document image can be confirmed. Bibliographic information such as the title of the document and the author's name corresponds to such information, which is also most suitable as a keyword for retrieval. In other words, when reading a document image with a scanner and saving it in the database, the logical structure of the document image is recognized to extract the bibliographic information,
It is useful to set this as a keyword.

[0010]

The invention according to claim 1 is
Image recognition means for inputting a document image, image division means for dividing the input document image into a plurality of elements according to preset conditions, and character recognition for recognizing characters from each of the plurality of divided elements Means is provided, area detection means for detecting a specific element as a bibliographic area is provided, and structure recognition means for recognizing the logical structure of the detected bibliographic area is provided,
The structure output means for outputting the recognized logical structure is provided .
The structure recognition means is a plurality of preset character string patterns.
Character strings in the bibliographical area as a logical structure
Recognize and match the priority of the predefined string pattern.
Accordingly, the recognition of the logical structure is regulated.

The bibliographical area referred to in the present invention is an area where bibliographical information such as a title and an author name is located. Specifically, a title description area located at the center of the upper edge of the first page, An area for writing the author's name, which is located in the lower right corner, etc. The logical structure is a set of information that can identify a document well, and specifically, a set of bibliographic items that are selectively extracted from the document and recognized up to the description content.

[0012]

[0013]

[0014]

[0015]

[0016]

[0017]

According to a second aspect of the present invention, in the first aspect of the present invention, the structure recognition means converts the character string in the bibliographical region that matches the character string pattern according to a preset conversion condition, and then determines the logical structure. recognize.

[0019]

According to the present invention, when a document image is input by the image inputting means, the input document image is divided into a plurality of elements by the image dividing means according to a preset condition, and the divided image is divided. A character is recognized by the character recognition means from each of the plurality of elements. When a specific element is detected as a bibliographic area by the area detecting means, the logical structure of the detected bibliographic area is recognized by the structure recognizing means, so that the structure outputting means outputs the logical structure of the document image. The structure recognition means uses a plurality of preset character strings.
Logical structure of the character string in the bibliographic area that matches the pattern
If the input accuracy of the document image is not good
However, the logical structure is accurately recognized, and a predefined character string
Regulate logical structure recognition according to pattern priority
So that multiple candidates for the logical structure are properly selected.
be able to.

[0020]

[0021]

[0022]

[0023]

[0024]

[0025]

According to the second aspect of the invention, the structure recognition means recognizes the logical structure after converting the character string in the bibliographical area that matches the character string pattern according to a preset conversion condition, so that misrecognition is predicted. It is possible to recognize the logical structure after appropriately modifying the character string.

[0027]

An embodiment of the present invention will be described below with reference to the drawings. First, as shown in FIG. 1, the document processing apparatus 1 of the present embodiment includes an image input unit 2, an image dividing unit 3, a character recognition unit 4, an area detection unit 5, a structure recognition unit 6, and a structure output unit 7. Have

The image input means 2 inputs a document image 8 which is an image of a document described on the document as image data of a dot matrix by scanning and scanning the document with an image scanner (not shown). The image dividing unit 3 divides the input document image 8 into a plurality of elements according to preset conditions, and the character recognizing unit 4
Recognizes a character from each of a plurality of divided elements.
Such element division and character recognition are realized by a known technique, and such a technique is disclosed in, for example, Japanese Patent Laid-Open No. 5-159101.

Although the input document image 8 may consist of multiple pages, the bibliographic information in question here is predicted to be located on the first page. Only executed on the page.

The area detecting means 5 detects a specific element as a bibliographical area among the plurality of elements divided as described above. More specifically, two conditions are set in advance, "the top edge of which is above the page" and "the left edge of which is on the left side of the page." To do. From the plurality of elements sorted in this way, an element whose width matches the effective area of the document image 8 is detected, and an element located above this element is detected as a bibliographical area.

That is, it is assumed that the bibliographic information such as the title and the author name is located above the body, and the width of the body is predicted to match the effective area of the document image 8. Therefore,
A wide element located above and to the left is detected as the upper portion of the text, and an element above this is detected as the bibliographical area.

The structure recognizing means 6 recognizes the logical structure of the bibliographical area detected as described above. More specifically,
Since bibliographical items such as titles and author names are described in the bibliographical region, the logical structure of the bibliographical region is recognized by extracting these bibliographical items individually. For this reason,
As shown in, the bibliographic item extraction rule is preset for each type, and various bibliographic items are extracted from the bibliographic area by this extraction rule.

In FIG. 2, the first field 9 is the type of bibliographic item, the second field 10 is the content of the extraction process, the third field 11 is the character string pattern, the fourth field 12 is the priority order, and the fifth field 13 is The contents of the conversion process are shown below. For example, the title is extracted by centering,
The priority is "1" and no conversion process is set.
The date is extracted by pattern matching with the character string pattern, and the priority order is set for the two character string patterns. A character string pattern with a priority of "2" corresponds to erroneous recognition of a character, and the beginning is the alphabet "l", and the conversion process corresponding to this is "l".
(L) → 1 (Ichi) ”.

The structure output means 7 outputs the logical structure of the document image 8 recognized as described above. The output format of this logical structure is, as shown in FIG.
al Organization for Standardization) Standard SGML
It is set as (Standard Generalized Markup Language) format. The meaning of each item in the SGML format is frontm ... Set of bibliographic items title ... Document title date ... Document date y ... Year m ... Month d ... Day key ... Other bibliographic items body. In this case, "frontm ~ key" is a bibliographic item extracted from the bibliographic area, and the image data is output after being converted into a character code, but "body" is a text area that is not the bibliographic area and the image The data is output without being converted to a character code.

With such a configuration, when the document image 8 is input to the image input means 2, the document processing apparatus 1 of this embodiment outputs the logical structure of the document image 8 from the structure output means 7. The processing operation of the document processing apparatus 1 will be described with reference to the flowchart of FIG.

First, the image input means 2 inputs the document image 8 as image data (step S1), and the input document image 8 is divided into a plurality of elements by the image dividing means 3 (step S2). The character recognition means 4 recognizes the characters of each of the plural elements thus created (step S3). Next, the area detecting means 5 sorts a plurality of elements in accordance with the above-described conditions, and then detects an element having a width that matches the effective area of the document image 8 as a body area (step S4). The element located above the element of the area is detected as the bibliographical area (step S
5).

Since there are bibliographic items such as "title, transmission date, transmission source, transmission destination" in the bibliographical area thus detected, these bibliographical items are recognized as a logical structure by the structure recognizing means 6. (Step S6). At this time, bibliographic items such as dates for which the character string patterns are predicted are extracted from the bibliographic area by matching with a plurality of character string patterns.

Since the character strings thus extracted are sorted according to the priority order of the character string patterns, when there are a large number of extracted character strings, only the character strings with a high priority order are adopted, and an increase in noise is prevented. To be done. Further, when the character string extracted as described above matches the conversion condition, the character string is subjected to conversion processing and then recognized in the logical structure. More specifically, the number "1" is often used for the date which is a bibliographic item, but this is often mistakenly recognized as "l" in the alphabet in character recognition. Since it is unlikely that this lowercase letter "l" is used as the first character of the bibliographic item, if it is located at the beginning of the bibliographic item, it is converted into "1".

When the logical structure of the document image 8 is recognized as described above, the structure output means 7 outputs the logical structure as shown in FIG.
The logical structure of the document image 8 is output in SGML format (steps S7 and S8). At this time, the bibliographic items extracted from the bibliographic area are converted into character codes and output, but the text area is output as image data.

As described above, the document processing apparatus 1 of this embodiment
Then, when the document image 8 is input to the image input means 2, the logical structure of the document image 8 is output from the structure output means 7. Therefore, even when the document image 8 is stored in the database, the keyword is determined by the logical structure. It can be set well. At this time, it is not necessary to detect the bibliographic region in which the bibliographical items are located from the document image 8 and set only this bibliographic region as the recognition target of the logical structure and to analyze the entire document image 8.
The logical structure is accurately recognized by a simple processing operation.

Further, since the bibliographic information discriminated by the logical structure is coded by the character recognizing means 4, even when the display for displaying the document image 8 read out from the database has a low resolution and the characters of the text cannot be read. Since the coded logical structure can be clearly read, the description content of the document image 8 can be confirmed. Moreover, since the body area which is not the bibliographical area is not converted into the character code, the processing operation load is reduced, the time is shortened, and the body of the document image 8 is accurately saved.

Further, in the document processing apparatus 1 of the present embodiment, the element of the body area is detected as an element whose width matches the effective area of the document image 8. Therefore, the element is detected with high accuracy by a simple processing operation. . Further, since the element of the bibliographical area is detected as an element located above the element of the text area, the element is detected with high accuracy by a simple processing operation. However, the present invention is not limited to the above-mentioned embodiment, and detects the elements of the text area as a number of elements in which the number of lines is larger than a preset reference value, and the number of characters is larger than a preset reference value. It is also possible to detect as many elements. That is, since the text generally has a large number of lines and characters compared with the bibliographical items, the text area and the bibliographical area can be detected with high precision by a simple processing operation by using this. In addition,
The processing operation in this case is easily realized by counting the number of lines or the number of characters of each of the plurality of elements divided from the document image 8 and comparing this with a reference value.

Further, in the document processing apparatus 1 of the present embodiment, since the character string of the bibliographical area that matches a plurality of preset character string patterns is recognized as a logical structure, various logical structures can be accurately recognized. You can Moreover, since the recognition of this logical structure is executed in accordance with the priority order of the character string pattern, the number of bibliographical items to be extracted is limited, and the character string that is likely to cause misrecognition is predicted and eliminated. You can do something like that. However, the present invention is not limited to the above-described embodiment, and the permissible range of the search of the document image 8 can be expanded by recognizing all the bibliographic items that match the character string pattern as a logical structure. is there.

Furthermore, in the document processing apparatus 1 of this embodiment,
Since the character string in the bibliographical area that matches the character string pattern is converted according to the preset conversion condition, the date "1" that is easily mistakenly recognized as the alphabet "l" is corrected as described above. Even if the input accuracy of the document image 8 is low and a malfunction occurs in character recognition, the logical structure of the document image 8 can be accurately recognized.

[0045]

According to the first aspect of the present invention, image input means for inputting a document image is provided, and image dividing means for dividing the input document image into a plurality of elements according to preset conditions is provided. Character recognition means for recognizing a character from each of the plurality of elements, area detection means for detecting a specific element as a bibliographic area, structure recognition means for recognizing the logical structure of the detected bibliographic area, and recognition Since the logical structure is output only from the bibliographical area rather than the entire document image by providing the structure output means for outputting the logical structure, the processing range is reduced and the load of the logical structure recognition processing is reduced. Since there are bibliographic items that can specify the document image well in the bibliographical area, the logical structure of the document image can be accurately recognized. The structure recognition means is a plurality of preset
Character strings in the bibliographical area by matching with character string patterns
By recognizing that it is a logical structure, various logical structures can be simplified.
It can be accurately recognized by easy processing operation and is specified in advance.
The logical structure is recognized according to the priority of the string pattern
By regulating knowledge, the number of recognized logical structures
In this case, this can be selected or weighted according to priority.
You can

[0046]

[0047]

[0048]

[0049]

[0050]

[0051]

In the invention according to claim 2 , the structure recognition means recognizes the logical structure after converting the character string of the bibliographical area that matches the character string pattern according to a preset conversion condition. Since a character string that is likely to be erroneously recognized can be predicted and corrected, the logical structure can be satisfactorily recognized even when the input accuracy of the document image is low.

[Brief description of drawings]

FIG. 1 is a block diagram showing a document processing apparatus according to an embodiment of the present invention.

FIG. 2 is a schematic diagram showing an extraction rule of bibliographic information.

FIG. 3 is a schematic diagram showing an output format of a logical structure.

FIG. 4 is a flowchart showing a processing operation of the document processing apparatus.

[Explanation of symbols]

1 Document processing device 2 Image input means 3 image segmentation means 4 Character recognition means 5 Area detection means 6 Structure recognition means 7 Structure output means 8 document images

─────────────────────────────────────────────────── ─── Continuation of front page (56) References JP-A-3-127169 (JP, A) JP-A-5-342326 (JP, A) PRU94-36 A study on conversion of document image into SGML document, IEICE Technical Report, Japan, September 22, 1994, Vol. 94, No. 242, pp. 73-80 A Construction Method of Knowledge Base for Document Image Structure Analysis, Journal of Information Processing Society of Japan, Japan, January 15, 1993, Vol. 34, No. 1, pp. 75-87 PRU93-125 Layout analysis based on pattern classification method, Technical report of IEICE, Japan, January 1994, Vol. 93, No. 431, pp. 71-78 PRU94-42 A method for interpreting the logical structure of a document, IEICE Technical Report, Japan, October 20, 1994, Vol. 94, No. 291, pp. 25-32 (58) Fields surveyed (Int.Cl. ⁷ , DB name) G06K 9/00-9/82 G06F 17/30 G06T 1/00

Claims (2)

(57) [Claims] 1. An image input means for inputting a document image, an image dividing means for dividing the input document image into a plurality of elements according to a preset condition, and a character is recognized from each of the plurality of divided elements. Character recognition means, area detection means for detecting a specific element as a bibliographic area, structure recognition means for recognizing the logical structure of the detected bibliographic area, and structure output means for outputting the recognized logical structure. Yes, and the
The structure recognition means uses a plurality of preset character string patterns.
The character string in the bibliographical area is recognized as a logical structure by matching with
Knowledge and correspond to the priority order of predefined character string patterns
A document processing device characterized by regulating the recognition of a logical structure . 2. A structure recognition means, document processing according to claim 1, wherein the recognizing the converted by the conversion condition set string bibliographic region matching the character string pattern in advance a logical structure apparatus.