JP3653156B2 - Document image area extraction method - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a document image region extraction method for accurately extracting a character region from a document without columns.
[0002]
[Prior art]
Various methods for extracting a text area from a document image have been proposed. For example, the character region extraction method previously proposed by the present applicant extracts a coherent character region by integrating small elements such as characters from a document image (see Japanese Patent Laid-Open No. 5-81475). . However, in this method, since the character rectangles are integrated by the same method regardless of multi-columns or no columns (single column), excessive division (unintegration) of regions occurs in a document without columns with a wide character interval. there is a possibility.
[0003]
Therefore, it is detected from the document image by designating "one-column set" or by the region division method and column type discrimination method (Japanese Patent Application No. 7-194399) proposed previously by the present applicant. It is necessary to determine the column type of the document image including one column, a plurality of columns, and a free column based on the blank part or ruled line, and to perform processing adapted to the one column. Further, in this method, when extracting small text areas by integrating the small areas in the row direction, in the case of a one-column set, the small areas are integrated even if they are far apart.
[0004]
[Problems to be solved by the invention]
However, it is considered that the notification sentence corresponds to one set, but the bibliographic items such as the creator and address at the top of the page are left and right and should not be integrated. Therefore, it is not possible to correctly extract a region by largely integrating only because it is a one-column set.
[0005]
As another method, the edge of the step is obtained from the peripheral distribution of the beginning position of the character string, and the same attribute (belonging column number: plural) is given to the character string belonging to the same step, and the upper and lower sides having the same attribute There is a method of dividing and extracting a character area from an image by integrating adjacent character strings (see Japanese Patent Laid-Open No. 1-183784).
[0006]
However, even with this method, it is impossible to perform control so that the text parts are integrated even if they are separated from each other and the bibliographic items are not integrated. Further, even in the case of a single column, it is not preferable that noise having a size similar to characters existing at the end of the page and the text are integrated and divided into the same region. Furthermore, since the projection is simply estimated to estimate the edge portion of the step, it is difficult to distinguish between noise that should not be integrated and characters such as bulleted numbers.
[0007]
The present invention has been made in consideration of the above-described circumstances, and an object of the present invention is to provide a document image region extraction method for accurately extracting a character region from a document without columns.
[0008]
[Means for Solving the Problems]
In order to achieve the above object, according to the first aspect of the present invention, there is provided a document image region extraction method for extracting a character region from a document image, and extracting the character region by integrating the character elements. Take the projection in the vertical or horizontal direction of the character elements at the left and right or upper and lower ends, find the minimum value of the frequency of the projection, and the blank part whose absolute value is smaller than the predetermined threshold is greater than or equal to the predetermined threshold The blank portion is detected as an end blank portion, and if the character element existence width at the end from the minimum value is larger than a predetermined threshold, the detected end blank portion is noise and character. The character area is extracted by referring to the position of the end blank portion, and is not an end blank portion that divides the region .
[0009]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings.
FIG. 1 shows the configuration of an embodiment of the present invention. 101 is an image input means for inputting a document image such as a manuscript, 102 is a connected component extracting means for compressing the image and extracting a circumscribed rectangle of the connected component, and 103 is a character element, a table element, a dividing line, and the like. Connected component classifying means for classifying into 104, edge blank detecting means for detecting edge blanks from the extracted character elements, and title candidate for detecting title candidates from those classified as character elements by the connected component classifying means 103 A detection means 106 is a character element integration means 107 for obtaining a character region by integrating character elements with reference to the edge blank portion obtained by the edge blank detection means 104 and the title candidate position obtained by the title candidate detection means 105. A data storage unit for storing input images and various types of information being processed, 108 is an overall control unit, 109 is a data communication path, and 110 is an input document with no columns. It is a means or means for determining specifies the.
[0010]
FIG. 2 shows a processing flowchart of the present invention. The processing operation of the present invention will be described with reference to FIG.
First, a document image is obtained by the image input means 101 (step 201). The image input means is a scanner, a fax, or the like, and an image may be obtained from another device via a network. Next, it is designated that there is no column (step 202). When the column is automatically determined, for example, the column type determination method of the above-mentioned application (Japanese Patent Application No. 7-194399) may be used. In this case, since the determination is made after the document image information is obtained as a processing procedure, a processing step of “no column setting determination” is inserted between step 204 and step 205.
[0011]
Next, the connected component extraction unit 102 compresses (reduces) the document image to about 1/8, and then extracts the connected component (step 203). By the compression processing described above, adjacent characters (character strings) become one connected component, and figures and the like are connected together and extracted. The connected component classification means 103 classifies the extracted connected components into character elements, tables, dividing lines, diagrams, and the like (step 204). At the time of classification, characteristics such as the size and aspect ratio of the element, the black pixel density, and the presence and position of ruled line elements in the connected components are used.
[0012]
The following processing is the main feature of the present invention.
First, the edge blank detecting means 104 detects a edge blank portion from the extracted character element (step 205). FIG. 3 is a diagram for explaining edge blank detection. FIG. 4 is a detailed flowchart of step 204 in FIG. The detection of edge blanks will be described with reference to FIGS. In FIG. 3, 302, 305, 306, 307, 308, 309, 310, 311 and 312 are character elements, and 301 and 303 are noises.
[0013]
If each character element is extracted as shown in FIG. 3, the projection of the frequency of the character element at the end is taken in the row (horizontal) direction and the vertical direction (step 401). That is, this frequency represents the number of character elements per unit width in the horizontal and vertical directions. In the example of FIG. 3, only the left end is shown. The same applies to the right end.
[0014]
The minimum value of the frequency information is obtained, and it is set as the edge blank portion obtained when the absolute value is lower than a predetermined threshold and the blank portion has a width equal to or larger than the predetermined threshold (step 402). In this case, (in FIG. 3, more left) end than the minimum value because the character elements exist width 318 is likely not a noise greater than a predetermined threshold, the detected end blank portion, noise and a character area This is excluded as not being an end blank part. As a result, an edge blank portion is detected (if the row direction is vertical, the whole process is rotated 90 degrees before the above-described processing is performed). The position of the detected edge blank portion is recorded and used for processing (step 207) for integrating character elements to be described later.
[0015]
Note that, as a method of detecting the blank portion described above, ruled line extraction using a white run may be performed instead of projection. FIG. 6 is a flowchart illustrating another process for detecting the end blank. The processing will be described with reference to FIG. A long vertical white run is extracted at the end, and this connected component is obtained. The run threshold is set as a long value, and the connected component exceeding this threshold becomes a part of the ruled blank portion (step 601).
[0016]
Next, based on the obtained width and length of the blank ruled line, it is determined whether or not it is a blank ruled line dividing the edge noise and the character area (step 602). The position of the blank ruled line obtained in this way is recorded and used for processing (step 207) for integrating character elements to be described later.
[0017]
Subsequently, the title candidate detection unit 105 extracts title candidates (step 206). FIG. 5 is a detailed flowchart of step 206. The title candidate detection process will be described with reference to FIGS.
[0018]
First, the uppermost character element is detected near the center of the entire page in a predetermined range from the top of the page (steps 501 and 502). In the example of FIG. 3, the character element 305 corresponds to this. The length obtained by adding the character elements adjacent to the left and right of this character element (up and down if the line direction is up and down) is measured, and when this length is larger than a predetermined threshold, it is detected as a title candidate (step 503). This position is recorded and used later when integrating character elements (step 207).
[0019]
Finally, the character element integration means 106 integrates the character elements into the area (step 207). This process basically integrates character elements adjacent in the line direction into lines and further integrates them into regions as in the method described in the above-mentioned publication (Japanese Patent Laid-Open No. 5-81475).
[0020]
Here, the difference from the conventional method is that an edge blank part is used and a title supplementary position is used. Referring to FIG. 3 as an example, when character elements are integrated into lines, if there are end blank portions between the elements, the integration conditions (such as the distance between elements and the degree of coincidence between the upper and lower sides) are tightened. As a result, the character element 301 and the character element 305 having a long inter-element distance are not integrated, and the character elements 303 and 307 or 308 having a low matching degree are not integrated. However, the character element 302 which is the itemized number is integrated with 306. As a result, it is possible to correctly distinguish and process the edge noises 301 and 303 that are difficult to distinguish by size alone and the character elements. In addition, since the area is not simply divided by the blank portion, the character element 302 is not isolated.
[0021]
Also, above the position of the title candidate 305 (the bibliographic item existence range), the conditions for integrating character elements into a line are strict. This prevents the character elements 309 and 310 from being mistakenly integrated. Furthermore, since the character elements 311 and 312 having a long interelement distance are below the title candidates, they are integrated even if they are separated. As a result, the character region can be correctly extracted even when the space between characters in the uncolumnar original is wide. The character regions finally extracted by the above processing are 313, 314, 315, 316, and 317.
[0022]
The present invention is not limited to the above, and can be realized by software. When the present invention is realized by software, as shown in FIG. 7, a general-purpose processing device including a CPU, a ROM, a RAM, a display device, a hard disk, a keyboard, a CD-ROM drive, etc. is prepared. The computer storage medium stores a program for realizing the document image region extraction function of the present invention.
[0023]
【The invention's effect】
As described above, according to the present invention, the existence range of the bibliographic items is identified in the manuscript without columns, so that it is possible to accurately separate the areas of the body part and the bibliographic item part. Become.
[0024]
According to the present invention, even when there is noise at the edge of the document, a blank area or a blank ruled line is detected, and the position is used when extracting the character area. Can do.
[Brief description of the drawings]
FIG. 1 shows a configuration of an embodiment of the present invention.
FIG. 2 shows a processing flowchart of the present invention.
FIG. 3 is a diagram for explaining edge blank detection;
FIG. 4 is a detailed flowchart of step 204 in FIG. 2;
FIG. 5 is a detailed flowchart of step 206;
FIG. 6 is a flowchart illustrating another process for detecting an edge blank.
FIG. 7 shows a configuration example when the present invention is realized by software.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 101 Image input means 102 Connected component extraction means 103 Connected component classification means 104 Edge blank detection means 105 Title candidate detection means 106 Character element integration means 107 Data storage part 108 Control part 109 Data communication path 110 Non-columnar designation means or determination means

Claims (1)

A document image region extraction method for extracting a character region from a document image and extracting the character region by integrating the character elements, wherein the character elements at the left and right ends or upper and lower ends of the document image are vertically or horizontally Taking a projection, the minimum value of the projection frequency is obtained, and when the blank portion whose absolute value is smaller than a predetermined threshold has a width equal to or larger than the predetermined threshold, the blank portion is detected as an end blank portion. When the character element existence width at the end from the minimum value is larger than a predetermined threshold value, the detected end blank portion is not an end blank portion that divides noise from the character region, and the end blank portion A character image region extracting method, wherein a character region is extracted with reference to a position of the character image.

Images