JPH1011459A - Document registration system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a document registration system which can automatically register an inputted document for retrieval together with a retrieval key. SOLUTION: A document recognition processing part 12 extracts an area having an attribute of a table, a figure, etc., and character information from an inputted document image. A document classification processing part 13 specifies an attribute of the format, etc., of the document image from the extracted area, and includes the specified document attribute, character information, etc., in the retrieval key and registers them in a document data base 14. At retrieval time, a retrieval key is inputted to an API from a WWW client terminal 18 through a LAN 17. According to the retrieval key, the API extracts the desired document image from the document data base 14.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to a document registration system for automatically registering a document image taken from a scanner or the like together with a search key in a database.

[0002]

2. Description of the Related Art In recent years, there has been an increasing demand for a filing system which digitizes a paper-based document and stores or manages the document in a computer. In a conventional filing system of this type, a document image (raster image) of a document is acquired through, for example, a facsimile or a scanner, and the document image is registered in a storage device in a computer as needed. At this time, a search key corresponding to the document image is input using a keyboard or the like when the document image is registered in consideration of the document search.

FIG. 16 is a diagram showing an example of a system screen displayed to allow a user to input a necessary search key when registering a document image. On this system screen, for example, each area such as a registration date, an author, a title, a genre, and a keyword is displayed. When necessary information is input to each area, the information is registered in a database together with a document image. It has become so.

[0004]

However, in the conventional document registration system, it is necessary to manually register a search key from a keyboard or the like while viewing a system screen as shown in FIG. 16 for each document image. However, there has been a problem that the burden on the registration work increases as the amount of the registered document image increases.

[0005] In addition, for example, when a keyword or the like is registered as a search key, if the keyword differs depending on the subjectivity of the user, there is a problem that the search efficiency is significantly reduced.

Accordingly, an object of the present invention is to provide an improved document registration system capable of automatically generating and registering an optimum search key from an input document image when registering the input document image. To provide.

[0007]

In order to solve the above-mentioned problems, the present invention comprises document registration means for registering an input document image in a database together with a search key corresponding to the document image, based on the search key. In the document registration system for searching for a registered document image, the document registration unit includes a document recognition processing unit that specifies attribute-based areas such as a character area, a table area, and a figure area on the document image and position information thereof. And automatically generating the search key including the attribute and position information of the specified attribute-based area.

More specifically, the document recognition processing section includes an image normalizing section for correcting the inclination of the document image when the document image is inclined, and extracting a basic rectangular area from the document image or the document image subjected to the inclination correction. A basic rectangular area extracting unit, an area integrating unit that integrates the extracted basic rectangular areas as necessary, and determines whether the basic rectangular area or the integrated area corresponds to a character, a figure, a table, or any other attribute. And a region attribute discriminating unit. The image normalization unit is configured to detect, for example, the peripheral distribution of each pixel in the row direction and the column direction in the document image, and rotate the document image so that the white distribution in each direction is maximized.

Further, the area attribute discriminating section obtains, for example, the variance of the width and height of the basic rectangular area included in the integrated area, the number of line segments in each area, and the length thereof. , The number of line segments, and the length of the line segments to determine whether the attribute of the area is a character, a table, or a figure.

The area attribute discriminating unit calculates, for example, a positional shift amount from an area adjacent to the target area determined to be a character. When the positional shift amount is equal to or less than a specified value, the target area is determined. The title display area is determined to be a neighboring area.

In another document registration system according to the present invention, the document registration means specifies a region by attribute such as a character region, a table region, and a drawing region on the document image and position information thereof, A document classification processing unit for performing classification in a document format based on the specified attribute-based area and its position information, wherein the classification result is included in the search key.

In such a document registration system, the document classification processing unit converts the document image into a character-based “character-based” based on, for example, the number of attributes of the attribute-based area recognized by the document recognition processing unit. It is configured to be classified into general document types such as "document", "table document" mainly composed of tables, and "graphic document" composed mainly of figures.

Alternatively, the type of the column of the document image is determined based on the existence frequency of the basic rectangular region or the integrated region in a predetermined direction and the function of the area of each region, and the document image is determined for each determined column. Is configured to be classified.

Alternatively, a feature vector generating means for generating a feature vector representing a document feature of a document image based on the attribute and position information of the attribute-based area recognized by the document recognition processing section; Is compared with the distance between the feature vector generated by the feature vector generating means and each feature vector stored in the dictionary unit, and a feature value having a smaller distance is compared. Group specifying means for specifying one group to which the vector belongs, and the document images are classified according to the specified group.

[0015]

Embodiments of the present invention will be described below in detail with reference to the drawings. FIG. 1 is a block diagram showing an embodiment in which the document registration system of the present invention is used by connecting to the Internet. The document registration system includes a server 1 for registering a document image and managing the document image so as to be searchable, and a LAN (local area
A plurality of WWWs connected via a network 17
(World Wide Web) and a client terminal 18.

The configuration of the server 1 is divided into a registration system and a search system. The registration system includes a scanner 11 for acquiring a document image of the document 10 for registration, a document recognition processing unit 12 for recognizing the document image of the acquired document 10, and classifying and responding to the document images recognized by the document recognition processing unit 12. And a document database 14 for storing classified document images and search keys corresponding to the document images. The search system is L
It comprises an httpd (hypertext transfer protocol daemon) 16 connected to the AN 17 and a database API (application program interface) 15 for searching. For convenience of explanation, it is assumed that the character image is a monochrome image.

First, the operation of the registration system of the document registration system having the above configuration will be described. First, basic functional blocks of the document recognition processing unit 12 will be described with reference to FIG. The document recognition processing unit 12 corrects the inclination of a document image captured from the scanner 11 during scanning to normalize the document image, and extracts a basic rectangular area from the normalized document image. A basic rectangle extraction unit 122, an area integration unit 123 that integrates the extracted basic rectangles, an area attribute determination unit 124 that performs attribute determination on each of the integrated areas, and an automatic character A character recognition unit 125 for performing a recognition process.
In the character recognition unit 125, for example, a commercially available OCR can be used.

The contents of the function blocks 121 to 125 will be described in more detail. 3 to 5 show the image normalization unit 121.
FIG. 4 is an explanatory diagram of the function of FIG. At the time of image normalization, the number of black pixels is detected by scanning the document image acquired by the scanner 11 in the x-axis direction (from left to right) and in the y-axis direction (from top to bottom). . FIG. 3 shows a document image 31 and
A peripheral distribution 32 of the number of black pixels obtained by scanning the document image 31 in the y-axis direction and a peripheral distribution 33 of the number of pixels obtained by scanning the document image 31 in the x-axis direction are shown.

When the document image 31 has a normal inclination (that is, is not inclined), as shown in FIG.
The white distribution (portion where the number of black pixels is 0) corresponding to the margins between the lines and the margins at the right end, the left end, the upper end, and the lower end in the peripheral distribution is clear. On the other hand, in the case of the inclined document image 41,
The distribution of white in the peripheral distribution 42 in the y direction and the peripheral distribution 43 in the x direction is relatively reduced as shown in FIG. As described above, the presence or absence of the inclination of the document image can be determined from the peripheral distribution in the x direction and the y direction. When the character image has no inclination, the white distribution in the peripheral distribution in the x direction and the y direction becomes maximum. ing.

In the image normalizing section 121, if
When the document image 41 is inclined as shown in (b), the document image 41 is rotated so as to maximize the white distribution in the peripheral distributions 42 and 43 in the x and y directions. FIG.
It is explanatory drawing of the detection method of the rotation angle which maximizes the white distribution in a peripheral distribution. As is clear from FIG. 5, in the present embodiment, from the relationship between the total length of white distributions in the peripheral distribution in the x direction and the y direction (Run Length of white) and the rotation angle, the length of the white distribution is maximum. Is detected. Then, the document image is rotated by the rotation angle t.

FIGS. 6A and 6B are explanatory diagrams of a basic rectangle extracting process in the basic rectangle extracting section 122. FIG.
3 is a pixel square area. In the illustrated example, the black pixel P
Eight pixels are P (i−1, j−1) and P (i, j) in the clockwise direction from the upper left of the square area around (i, j).
-1), P (i + 1, j-1), P (i + 1, j), P
(I + 1, j + 1), P (i, j + 1), P (i-1,
j + 1) and P (i-1, j). The basic rectangle extraction unit 122 determines whether the eight pixels surrounding the black pixel P (i, j) are black or white.
For each pixel determined to be a black pixel,
A determination is again made as to whether or not the pixel newly determined to be a black pixel is black with respect to eight pixels surrounding the pixel. This operation is repeated until no new black pixel is found.
FIG. 6 shows an example of a rectangular area obtained after one operation is completed.
(B).

Each of the extracted basic rectangles is stored in the area integrating unit 12.
3 is integrated. As the integration logic, FIG.
7 and the logic shown in FIG. 7B can be applied. In the logic of FIG. 7A, attention is paid to a certain rectangular area Ci, and the rectangular area Ci
A fixed distance xd outward from the pair of long and short sides, respectively
An adjacent space is set in a range separated by ist and ydist. The distances xdist and ydist are distances corresponding to the resolution of the document image. Then, another rectangular area Cj in the close space is detected, and an integrated area Ck1 including the rectangular area Cj is determined. On the other hand, in the logic of FIG. 7B, another rectangular area Cj that at least partially overlaps or includes a certain rectangular area Ci is obtained by focusing on a certain rectangular area Ci, and an integrated area Ck2 including these is determined. The area integration unit 123 performs area integration by applying any of the above logics to all the basic rectangular areas extracted by the basic rectangle extraction unit 1 in the preceding stage.

Each integrated area is assigned to an area attribute discriminating unit 12.
At 4, the attribute is determined. The processing procedure in the area attribute discriminating unit 124 is as shown in FIG. 8, and first, a line segment in the area is extracted for each integrated area (S1).
1). Next, the variance σ of the width w and the height h of the basic rectangle in the area
_w ² and σ _h ² are calculated according to the following equations (1-1) and (1-2), and the respective variances σ _w ² and σ _h ² are determined by thresholds w _var and h
_It is determined whether it is smaller than _var (S12). In the equation below, N is the total number of the rectangular region, the w _a mean value of the width of the rectangular area, h _a is the mean value of the height of the rectangular area.

[0024]

(Equation 1)

If each of the variances σ _w ² and σ _h ² is smaller than the threshold value (S12: True), the length of the extracted line segment is 30% or more of the width or height of the region. It is determined whether or not a line segment exists (S13). If it does not exist (S13: False), the area is determined to be a character area (S14), and a figure title identification process is executed (S1).
5). On the other hand, when at least one of the variances is equal to or greater than the threshold in the determination step of S12 (S12: False
e) or each variance is smaller than the threshold (S1
2: True), if there is no extracted line segment whose length is 30% or more of the width or height of the region (S1)
3: True), the area is determined to be a table area or a figure area, and the process proceeds to S16.

In the process of S16, a table area and a figure area are determined. That is, of the extracted line segments, 80% of the width of the area
It is determined whether there are two or more objects having the above length and two or more objects having a length of 80% or more of the height of the region. If it exists (S16: True), the area is determined to be a table area (S17).
16: False) The area is determined to be a figure area.

Here, the contents of the figure title identification processing (S15) will be described with reference to FIG. In the figure title identification processing, FIG.
As shown in the figure, the area Ci (rectangular upper left coordinate: xi, yi) determined to be a character area by the area attribute determining unit 124 is focused on, and a neighboring rectangular area Cj (rectangular upper left coordinate: xj, y) is noted.
For j), the distance ydist in the y direction and the center displacement cdist are calculated. Then, figure title identification processing is performed by the following equation. In the following equation, wi, wj, h
i and hj represent the width and height of the rectangular areas Ci and Cj, respectively.

[0028]

(Equation 2)

When the above determination formula is satisfied, if the rectangular area Cj is a figure area, the rectangular area Ci is determined to be a character area for the title. If the rectangular area Cj is a table area, the rectangular area C is determined.
i is determined to be a character area for the title.

As a result of the above processing, a document image as shown in FIG. 10 is obtained. From FIG. 10, it can be seen that the text area is mainly divided into a character area mainly composed of characters, a figure area mainly composed of figures, and a table area mainly composed of tables. Further, a character area corresponding to a title or a title exists near the figure area or the table area.

Next, the operation of the document classification processing section 13 will be described. The document classification processing unit 13 includes the document recognition processing unit 12
Classification is performed based on the document format (format) based on the processing result. More specifically, the document images are classified into “character documents” mainly composed of characters, “table documents” mainly composed of tables, and “graphical documents” mainly composed of figures. This is called a general document type classification. In addition, it is classified into a two-column system, a three-column system, and the like found in magazines and articles. This is called a column classification.
Further, the data is classified into formats defined by the user according to the purpose. This is called a user-defined classification.

First, the general document type classification will be described. The document classification processing unit 13 calculates the area of the table area Cti as St
i, when the area of the graphic region Cfi is Sfi and the area of all the regions is S, it is determined whether the document is a table document, a graphic document, or a character document by the following determination formula.

[0033]

(Equation 3)

Next, the column classification will be described. For column classification, first, the frequency distribution in the x direction of each rectangular area,
That is, the existence frequency of the origin (for example, the upper left of the region) of the rectangular region in the x direction is checked, and the class value (representative value within a certain range) Ri in the x direction is multiplied by the product ni of the frequency ni and each region area Si.
The relationship with * Si (ni * Si = g (Ri)) is obtained.
This is called a g function. Next, the g function is multiplied by a filter f3 for three stages and a filter f2 for two stages to obtain a composite function h3 (Ri) = f3 (g (Ri)), h2
(Ri) = f3 (g (Ri)) is calculated. This is called an h function.

FIG. 11A shows the characteristics of a two-stage filter, FIG. 11B shows the characteristics of a three-stage filter, FIG. 11C shows the g function in this embodiment, and FIG. 11D shows the h function. FIG. 11C and 11D, the vertical axis represents the product ni * Si of the frequency ni and the area of each region Si, and the horizontal axis represents the coordinates in the x direction. The reason for multiplying each frequency ni by the area area Si is to give weight to each frequency ni and eliminate dust printing or the like in the document image. The h function shown in FIG. 11D is, after all, a product of the frequency distribution in the x direction and the area when the light passes through the filter. Therefore, the type of the column is determined based on the ratio of the h function to the total area S of the entire area (integrated area). The determination of the type of the column is made based on whether or not the following condition is satisfied.
Note that the order of determination is performed in the order of Expression (4-1) and Expression (4-2).

[0036]

(Equation 4)

Next, the user-defined classification will be described.
FIG. 12 is an explanatory diagram of a registration procedure of a format uniquely defined by a user. Specifying a single-attribute rectangular area for a document image input from the scanner 11 is the same as the above-described processing of the document recognition processing unit 12 (step S2).
1). Here, a feature vector as shown in FIG. 13 is extracted for the specified rectangular area (step S2).
2). Then, a dictionary for format determination is created from the feature vector (step S23). The format determination dictionary can be included in, for example, the document database 14, but may exist as an independent unit. The feature vector V shown in FIG.
1, V2,..., V53) (53 is the number of components), and each component of the feature vector represents a feature of the entire rectangular area specified during the document recognition processing.

Next, a processing procedure for determining a category of a format defined by a user using a format determination dictionary will be described with reference to FIG. The category classification of the format defined by the user is performed by the document classification processing unit 13.
First, the document classification processing unit 13 performs a document recognition process on the document image of the document input to the scanner 11 to specify a rectangular area having a single attribute (step S31). Next, based on the specified rectangular area, a feature vector V = (V1, V2,..., V53: 53 is the number of components) shown in FIG.
Is extracted (step S32). Further, a transformation matrix M is formed which forms a feature space in which feature vectors of documents having different format categories are more scattered and feature vectors of documents belonging to the same category are more closely distributed. New feature vector V
! = MV is defined (step S33).

FIG. 15 shows an example of a feature space. Points in the figure represent feature vectors for one document. Further, a set of point sets close to each other is classified into categories A, B, and C to which a format defined by the user belongs.
It corresponds to. Thus, the feature vector V! Is mapped to the feature space, and the Euclidean distance from the feature vectors included in the categories A, B, and C is calculated (step S34). Feature vector V! As an example of the definition of the distance between a category and a category, the feature vector V! The average value or the minimum value of the distances between the feature vector and the feature vector belonging to the category can be taken.

Finally, if a category having the closest distance to each category and a value smaller than the threshold value d is found, the format of the input document is determined to belong to the category. (Step S
35). If the distance between the format of the input document and the existing category both exceeds the threshold value d, the category of the input document is registered as a new category.

As described above, in the document registration system according to the present embodiment, the format category of the document defined by the user input from the scanner 11 is automatically determined. Then, the document image of the input document is registered in the document database 14 together with a search key including identification information on the determined category. A search of the document database 14 is executed based on the search key.

Next, the operation of the retrieval system of the document registration system will be described. Document search is, for example, a WW connected to a LAN or the like.
This is performed by inputting a search key from the W client terminal 18. Search application is a commercial WW
A W browser can be used. Also, httpd
A more search interface is provided. Then, a desired document image is retrieved from the document database 14 through the database API using, for example, “cgi-bin”. In this way, a search can be executed interactively from the WWW client terminal 18 to obtain a document desired by the user. The search result in this case is generally a plurality of documents. Further, when the number of documents in the search result is reduced in consideration of the number of documents in the search result, a document image can be output and referenced at a stretch by browsing (parallel flipping). The retrieved document is displayed on the display of the WWW client terminal 18 or output by printing out.

In this document registration system, three types of searches by format search, keyword search, or figure / table name search are possible. In the format search, the "character document", "table document", "graphic document",
The document is searched by referring to the format information of the document registered in the document database 14 using “column-less”, “two-column”, “three-column” or a format defined by the user as a search key. . In the keyword search, a search is performed with reference to character information of a document registered in the document database 14 using a keyword as a search key. At this time, a commercially available full-text search engine may be used. Also,
In the figure / table name search, a search is performed using the figure / title given to the figure as a search key. The figure / table name information of the document registered in the document database 14 is referred to, and the designated title is designated. Search for documents in which figures / tables containing "" are described.

Although the document registration system according to the present embodiment has a form connected to a wide area network such as the Internet, such a form is not necessarily required. For example, an embodiment in which a scanner for taking in a document into a personal computer programmed to realize document registration and document search and a display or printer for outputting a search result are connected is also possible. As described above, the present invention has been described with reference to the embodiments. However, it is needless to say that the present invention is not limited to the above embodiments.

[0045]

As is apparent from the above description, according to the present invention, the burden of inputting and classifying document images to be searched can be reduced, and high-precision searching can be performed. Therefore, a practical filing system can be constructed and operated, and paperless office work is promoted, so that it is not necessary to secure a document storage space.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a document registration system according to an embodiment of the present invention.

FIG. 2 is a configuration diagram of basic functional blocks of a document recognition processing unit according to the embodiment.

FIG. 3 is a peripheral distribution diagram of black pixels existing on a document image according to the embodiment;

4A is a peripheral distribution diagram of black pixels existing on a document image having no inclination, and FIG. 4B is a peripheral distribution diagram of black pixels existing on an inclined document image.

FIG. 5 is an explanatory diagram of calculating a tilt correction angle of a document image according to the embodiment.

FIG. 6A is a diagram illustrating a process of extracting a basic rectangular area according to the embodiment, and FIG. 6B is a diagram illustrating an example of the basic rectangular area according to the embodiment.

7A is an explanatory diagram of integrated logic of a basic rectangular area according to the embodiment, and FIG. 7B is an explanatory diagram of another integrated logic of the basic rectangular area according to the embodiment.

FIG. 8 is an explanatory diagram of a processing procedure in a region attribute determination process according to the embodiment;

FIG. 9 is an explanatory diagram of a processing procedure in a figure title identification processing according to the embodiment;

FIG. 10 is a view showing an example of a document image after the document recognition processing according to the embodiment.

11A is a characteristic diagram of a two-stage filter according to the present embodiment, FIG. 11B is a characteristic diagram of a three-stage filter, FIG. 11C is a diagram showing an example of a g function, and FIG. The figure which showed an example of the h function (synthesis function) in the case of c).

FIG. 12 is an explanatory diagram of a processing procedure in a user-defined format registration process according to the embodiment;

FIG. 13 is a view showing an example of the definition of a feature vector according to the embodiment.

FIG. 14 is an explanatory diagram of a processing procedure in a format determination process according to a user definition according to the embodiment;

FIG. 15 is a view showing an example of a feature space according to the embodiment;

FIG. 16 is a diagram showing an example of a registration screen in a conventional document registration system.

[Explanation of symbols]

 1 Server 10 Input Document 11 Scanner 12 Document Recognition Processing Unit 13 Document Classification Processing Unit 14 Document Database 15 API 16 http 17 LAN 18 WWW Client Terminal 121 Image Normalization Unit 122 Basic Rectangle Extraction Unit 123 Area Integration Unit 124 Area Attribute Discrimination Unit 125 Character recognition unit

 ──────────────────────────────────────────────────の Continuing on the front page (72) Inventor Hideaki Tsukamoto 3-3-3 Toyosu, Koto-ku, Tokyo NTT Data Communications Corporation

Claims (9)

[Claims] 1. A document registration system comprising: document registration means for registering an input document image in a database together with a search key corresponding to the document image; and searching for a registered document image based on the search key. The document registration unit has a document recognition processing unit that specifies attribute-based areas such as a character area, a table area, and a figure area on the document image, and position information thereof, and stores the attribute and position information of the specified attribute-based area. A document registration system for automatically generating the search key. 2. The image recognition processing unit according to claim 1, wherein the document image processing unit corrects the inclination of the document image when the document image is inclined, and extracts a basic rectangular area from the document image or the document image subjected to the inclination correction. A basic rectangular area extracting unit, an area integrating unit that integrates the extracted basic rectangular areas as necessary, and an area for determining whether the basic rectangular area or the integrated area corresponds to a character, a figure, a table, or any other attribute. 2. The document registration system according to claim 1, further comprising an attribute determination unit. 3. The image normalizing unit detects a peripheral distribution of each pixel in a row direction and a column direction in the document image, and rotates the document image such that a white distribution in each direction is maximized. The document registration system according to claim 2, wherein 4. The area attribute determining unit obtains the variance of the width and height of the basic rectangular area included in the integrated area, the number of line segments in each area and the length thereof, and calculates the value of each variance. , The number of line segments, and the attribute of the region according to the length of the line segments, characters, tables,
3. The document registration system according to claim 2, wherein one of the figures is determined. 5. The region attribute determining unit calculates a positional shift amount of a region of interest determined as a character from a region near the target region. 5. The document registration system according to claim 4, wherein the region is determined to be a title display region. 6. A document registration system, comprising: document registration means for registering an input document image in a database together with a search key corresponding to the document image, wherein the document registration system searches for a registered document image based on the search key. The document registration unit includes: a document recognition processing unit that specifies an attribute-based area such as a character area, a table area, and a figure area on the document image and its position information; and a document based on the specified attribute-based area and its position information. A document registration system, comprising: a document classification processing unit for performing classification by format; and including the classification result in the search key. 7. The document classification processing unit, based on the number of attributes in the attribute-based area recognized by the document recognition processing unit, the document image is a “character document” mainly composed of characters, and a “table document” composed mainly of tables. 7. The document registration system according to claim 6, wherein the document registration system is classified into general document types such as "graphic documents" and "graphic documents" mainly composed of graphics. 8. The document classification processing unit determines the type of a column of the document image based on a function of the frequency of existence of the basic rectangular area or the integrated area in a predetermined direction and the area of each area. 7. The document registration system according to claim 6, wherein the document images are classified according to columns. 9. A feature vector generation unit configured to generate a feature vector representing a document feature of a document image based on attribute and position information of an attribute-based area recognized by the document recognition processing unit, A dictionary unit in which a plurality of feature vectors generated in advance are grouped for each document format and stored, and a distance between a feature vector generated by the feature vector generation unit and each feature vector stored in the dictionary unit are compared. 7. The document registration system according to claim 6, further comprising group specifying means for specifying one group to which the feature vector having the smaller distance belongs, and classifying the document image according to the specified group.