JP2968284B2 - Character recognition apparatus and character area separation method - Google Patents

Description

The present invention relates to a character recognition device for recognizing characters from binary image data (image information) obtained by an image scanner or other optical reading device. Further, the present invention relates to a character area separating method for separating and extracting a character area and a graphic area which are pre-processing of character recognition in the character recognition apparatus.

[Conventional technology]

In the character recognition process for recognizing a character from binary image data, image information (character image) corresponding to each character is obtained by performing a predetermined extraction process from the image data in the character area.
However, in the case of a document containing a figure, it is essential to separate the character area and the figure area as preprocessing.

2. Description of the Related Art Conventionally, in a character recognition apparatus, separation of the character area and the graphic area (hereinafter, referred to as “character area separation method” for simplicity) is performed by first processing binarized (black and white) image data. There is a method in which histograms are alternately arranged vertically and horizontally, and a continuous area is treated as one area to perform area division. However, in this method, a desired restriction necessary for the area division is provided, and the same processing is repeated until the restriction is satisfied. Therefore, when the processing amount is large and the number of black areas is large, adaptation may be difficult.

There is also a method of labeling the multi-valued image data by dividing it into small areas, but the multi-valued image data has a larger amount of information than the binarized image data and requires a large-capacity image memory. Had become.

Also, in the method of dividing binary image data into small regions and labeling them, when the small region is large, the processing speed is fast but the resolution is poor, and when the small region is small, the resolution is good but the processing is good. There was a conflicting problem that the speed was slow.

By the way, a method of tracing the outline of a black area (black pixel) to determine an area where the black area exists, treating the area in a rectangular (box) expression, and separating the character area from the graphic area using the height as a parameter. Proposed. This method
High-resolution processing can be performed in a relatively short time.

[Problems to be solved by the invention]

In other words, this method makes it easy to separate the character area from the graphic area according to the size (threshold) of the circumscribed rectangular area of the contour, and furthermore, for example, a character area surrounded by a graphic like a character in a table. However, there is a tendency that each character region is separated into fine regions as the resolution increases.

On the other hand, in the character area subjected to the character recognition processing, character spacing, line spacing and other properties between characters are accurately grasped,
In order to stably perform the cutout processing for each character, a certain number of characters are required in the area.

However, in the above-mentioned method, when the separated and extracted character area is small, the clipping process for obtaining individual character images becomes unstable, which may cause a reduction in the character recognition rate.

An object of the present invention is to provide a character recognition device and a character region separation method which can accurately separate and extract a character region, and can combine character regions into relatively large regions and avoid the conventional problems. With the goal.

[Means for Solving the Problems] In the character recognition device according to claim (1), the integration processing unit performs processing for each rectangular area separated as a character area.
A process of integrating adjacent rectangular regions and a process of removing “nesting” and “overlap” of the integrated rectangular regions are sequentially repeated, and the obtained rectangular regions are extracted as character regions.

In the character recognition device according to claim (2), in the character recognition device according to claim (1), the process of integrating the adjacent rectangular regions is performed based on a height histogram of each of the separated rectangular regions. The maximum value is obtained, and a plurality of rectangular areas whose distance between the rectangular areas is smaller than the maximum value are integrated to form one rectangular area.

In the character area separation method according to the third aspect, a rectangular area is extracted by performing contour tracing of a black area of the binary image data, and a histogram of the height of each rectangular area is obtained.
Further, a rectangular area as a character area is extracted based on the histogram, and the rectangular areas in which the distances between the extracted rectangular areas are relatively close are combined into one rectangular area. Is performed, and if there is a “nest” in which a rectangular area is present in the rectangular area or an “overlap” in which a plurality of rectangular areas are overlapped, the “nest” is removed. And the "overlap" removal process is performed, and the integration process and the removal process are repeated in order until there is no change in the rectangular region, and the obtained rectangular region is separated and extracted as a character region.

In the character area separating method according to the present invention, a rectangular area is extracted by tracing the contour of a black area of the binarized image data, a histogram of the height of each rectangular area is obtained, and further based on the histogram. Extracting a rectangular area as a character area by performing thickening processing with a predetermined thickness to the outside of the extracted rectangular area as a character area, and further performing thinning processing to reduce a predetermined thickness from the outside. In this process, rectangular areas closer to each other are connected to each other, and for each rectangular area obtained by the connection, the rectangular areas whose distances between the rectangular areas are relatively close are combined into one rectangular area. Performing the integration process, the rectangular area
If there is a “neft” where a rectangular area exists in the rectangular area or an “overlap” where a plurality of rectangular areas overlap,
The removal processing of “neft” and “overlap” for removing them is performed, and the integration processing and the removal processing are repeated in order until there is no change in the rectangular area, and the obtained rectangular area is separated and extracted as a character area.

(Operation)

According to the present invention, for each rectangular area separated as a character area, integration processing means integrates adjacent rectangular areas, and processing for removing “nest” and “overlap” of each integrated rectangular area. Can be combined into a larger rectangular area.

Therefore, by separating and extracting this large rectangular area as a character area, it is easy to stabilize the cutout processing in the subsequent character recognition processing.

Further, in the present invention, by performing boldening processing and thinning processing of each rectangular area separated as a character area or each rectangular area in the course of the integration processing, the integration processing efficiency is greatly improved.

〔Example〕

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 1 is a block diagram showing an example of an apparatus configuration for realizing the character area separation system of the present invention.

In the figure, the character recognition device 21 is activated together with the image scanner 25 in accordance with the activation control of the host device (personal computer) 23. The binary image data (image information) input from the image scanner 25 is stored in the corresponding memory (RAM) 27 of the character recognition device 21.

The character area separating unit 28 of the character recognition device 21 separates and extracts a character area using the memory 27 including the image memories I, II, and III and the tables I and II. Here, the separated character area is passed to a character recognition processing unit 29, a character extraction process is performed from the character area to obtain each character image by a predetermined cutout processing, and the character recognition processing is performed, and the recognition result is notified to the host device 23. is there.

FIG. 2 is a flowchart showing the procedure of an embodiment of the character area separation method of the present invention. FIG. 3 is a diagram showing an example of a processing result in each processing step.

Hereinafter, the flow of the separation extraction distance of the character area will be described with reference to FIGS.

The binarized image data (image information) read by the image scanner 25 is stored in the image memory I. This image data is converted into a 1/8 reduced image (FIG. 3 (a)) using an existing reduction algorithm, and is converted into an image memory.
Store in II.

On the reduced image (image memory II), the contour of a black area consisting of black points is traced, and a rectangular area (FIG. 3 (b)) represented by the minimum and maximum values of the coordinates of points on the outline of the black area. )), And the coordinate value data is stored in Table I. Note that the image memory III
Use

A histogram of the height of the rectangular area (FIG. 3 (c)) is obtained from the coordinate value data stored in the table I, and a threshold value for separating the graphic area from the character area (the height of a point beyond one peak, 3) in Fig. 3 (c).
Store in Table II. Subsequently, a rectangular area smaller than the threshold is extracted as a character area and stored in the table I.

The above processing is the same as the conventional method, and the following processing has features of the method of the present invention.

The contents of each rectangular area stored in the table I are painted out (FIG. 3D) and stored in the image memory II.

Subsequently, in order to connect adjacent rectangular areas, a bolding process is performed with a thickness of 2 dots to the outside of each rectangular area stored in the image memory II (FIG. 3 (e)). A thinning process for removing two dots is performed (FIG. 3 (f)) and stored in the image memory II. That is, by performing the thick line processing and the thin line processing, it is possible to fill a fine interval between adjacent rectangular areas.

Here, the contour tracing of the black area consisting of black points is performed again on the image memory II, a rectangular area (FIG. 3 (g)) in contact with the black area is similarly extracted, and the coordinate value data is stored in the table I. .

Note that the area separated and extracted here is a character area. However, since an area of one area is often separated into another area on a document, a relatively close rectangular area is integrated. Perform processing.

In the process of integrating the adjacent rectangular areas, the maximum value (☆ in FIG. 3 (c)) is obtained from the histogram of the height of each rectangular area stored in Table II, and the distance between the rectangular areas is calculated as follows. A process of integrating two small rectangular areas into one rectangular area is performed (FIG. 3 (h)) and stored in the table I. In this integration processing, a “nest” in which a rectangular area is present in a rectangular area, or an “overlap” in which a plurality of rectangular areas are overlapped occurs.

Therefore, if there is a "nest", a process to leave the larger rectangular area is performed, and if there is an "overlap", a new rectangular area including them is created and the old rectangular area is deleted. The "overlap" is removed (FIG. 3 (i)) and stored in Table I.

Hereinafter, the integration processing of the adjacent rectangular areas and the processing of removing the “nest” and “overlap” are repeated until the table I becomes steady (there is no change), and finally the rectangular area surrounding the character area (FIG. 3 (j )), The coordinate values of each rectangular area are stored in Table I, and the character area separation / extraction process ends.

The above processing is a feature of the method of the present invention,
The coordinate values of each character area obtained here are subjected to character recognition processing, and image cutout processing and recognition processing are performed for each character area.

〔The invention's effect〕

As described above, according to the present invention, it is easy to accurately separate a graphic region and a character region, for example, a character region in a table can be separated and extracted, and a character region that can be fully automated in character recognition processing can be obtained. In the separation method, each separated character area can be grouped into a large area, so that it is easy to stabilize the cutout processing corresponding to each character in the character area, and to improve the character recognition rate. it can.

In addition, by performing the thick line processing and the thin line processing of each character area before or during the integration processing, the integration processing efficiency can be greatly improved, and the memory capacity required for the integration processing can be reduced. it can.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an example of an apparatus configuration for realizing the method of the present invention, FIG. 2 is a flowchart showing the procedure of an embodiment of the method of the present invention, and FIG. 3 is a diagram showing an example of a processing result in each processing step. . In the figure, 21 is a character recognition device, 23 is a higher-level device (personal computer), 25 is an image scanner, 27 is a memory (RAM), 28 is a character area separation unit, and 29 is a character recognition processing unit.

Claims (4)

(57) [Claims] 1. A character recognition device that separates a character region from binary image data and recognizes a character, comprising: integrating a rectangular region adjacent to each rectangular region separated as a character region; A character recognition device, comprising: integrated processing means for sequentially repeating a process of removing “nest” and “overlap” of each integrated rectangular region and extracting an obtained rectangular region as a character region. 2. The character recognition apparatus according to claim 1, wherein the processing of integrating the adjacent rectangular areas includes obtaining a maximum value from a height histogram of each of the separated rectangular areas, A character recognition device, wherein a plurality of rectangular areas having a distance between them smaller than the maximum value are integrated into one rectangular area. 3. A character area separating method for a character recognition apparatus for performing character recognition from binary image data, wherein a rectangular area is extracted by tracing a contour of a black area of the binary image data. Is obtained, and a rectangular region as a character region is extracted based on the histogram. The rectangular regions in which the distance between the rectangular regions as the extracted character region is relatively short are put together. When one rectangular area is subjected to integration processing, and the rectangular area subjected to the integration processing has a “nest” in which a rectangular area exists in the rectangular area or an “overlap” in which a plurality of rectangular areas overlap, Performing a removal process of “nest” and “overlap” to remove them, and repeating the integration process and the removal process in order until there is no change in the rectangular area. Character segmentation method characterized by separating and extracting the rectangular area as a character area. 4. A character region separating method for a character recognition device for performing character recognition from binary image data, wherein a rectangular region is extracted by performing contour tracing of a black region of the binary image data. Is obtained, a rectangular area as a character area is extracted based on the histogram, and a thick line processing is performed with a predetermined thickness outside the rectangular area as the extracted character area. The adjacent rectangular areas are connected by performing a thinning process for reducing a predetermined thickness from the above. 1
In the case where there is a “nest” in which a rectangular area exists in the rectangular area or an “overlap” in which a plurality of rectangular areas overlap, Performing nesting and overlap removal processing to remove them, repeating the integration processing and the removal processing in order until there is no change in the rectangular area, and separating and extracting the resulting rectangular area as a character area. Characteristic region separation method.