JPH02159690A - Device for deciding character/pattern area - Google Patents

Abstract

PURPOSE:To speed up a processing and surely decide a character area from a pattern area even when characters and patterns are completely mixed by chain-coding a contour line, counting up the number of times of changing of the directions of the chained-code and counting up also the number of black picture elements. CONSTITUTION:A general document is digitally inputted from an image input part 1, its original image is stored in an image memory 2, the contour of the original image is extracted by a contour extracting part 3 and chain-coded by a chain-coding part 4 while tracing the contour line. Then, a direction change counting part 5 counts up the number of changes in the direction of the chained- code, a black connection counting part counts up the number of black picture elements in the original image connected to a series of chained codes and an area deciding part 6 decides a character area and a pattern area by utilizing the number of times of changing of the chained code and the area of the black picture elements. Since a large character also can be separated as a character and a small graphic also can be separated as a pattern, characters and patterns can be separated even if they are complexedly mixed.

Description

Detailed Description of the Invention (Field of Industrial Application) The present invention is applicable to document image recognition for recognizing general documents that are not marked for recognition by humans. The present invention relates to a device for identifying character areas and graphic areas in a document.

(Prior Art) Character/graphic area determination processing that is generally performed is based on (1) the size of a black pixel group, and (2) the density of neighboring lines.

(3) It uses peripheral distribution characteristics.

(Problem to be Solved by the Invention) However, in the above-mentioned method (1) of performing separation processing based on the size of the black pixel group, it is not possible to perform separation processing based on the size of the black pixel group alone. There is a problem that erroneous separation occurs.

Method of separating according to the density of the neighboring lines (2)
First, in order to check the density of neighboring lines, it is necessary to extract the strokes of the lines and examine their distribution, so the processing for this is complicated and generally takes a long time. There is a problem.

In method (3) using the above marginal distribution features,
In the marginal distribution, since black pixels are projected and added, there is a problem in that if characters and figures are not separated when viewed vertically and horizontally, it is difficult to separate them.

The present invention aims to solve these problems of the prior art. In other words, the aim is to be able to separate large characters as characters and small figures as figures, to speed up the process through relatively simple processing, and to be able to separate characters and figures even when there is a complex mixture of letters and figures. That is.

(Means for Solving the Problems) The structure of the present invention for achieving the above object is the first r! !
As shown in the embodiment of J, an image input section 1 that digitizes and inputs an image, an image memory 2 that stores the image, a contour extraction section 3 that extracts a contour in the image, and a contour extraction section thereof. a chain encoding unit 4 that traces the contour extracted in step 3 and chain encodes it; a direction change counting unit 5 that counts the number of changes in the direction of the chain code;
A black connection counting section 6 counts the black pixels connected to the chain code of the series, and the character area is calculated based on the number of direction changes counted by the direction change counting section 5 and the number of black pixels counted by the black connection counting section 8. and a graphic region.
It is characterized by having the following.

(Function) The present invention digitally inputs a general document (a document that has not been marked for human processing, etc.) from the image input unit l, and stores the original image in the image memory 2. The contour extraction section 3 extracts the contour of the image from the original image, and while tracing the contour, the chain encoding section 4 performs chain encoding, and the direction change counting section 5 calculates the direction of the chain code. The black connection counting unit 6 counts the black pixels of the original image that are connected to the series of chain codes (that is, the area of the black pixels surrounded by the series of chain codes), and the area determination unit 7 counts the black pixels of the original image that are connected to the series of chain codes. , character areas and graphic areas are determined using the number of changes in the chain code, the number of changes in the series of chain codes, and the area of black pixels surrounded by the series of chain codes.

(Embodiment) FIG. 1 shows an embodiment of the present invention, in which an image input section 1
．． Image memory 21 contour extraction section 3. Chain encoding unit 4j direction change counting unit 5. Black connection counting unit 61 area determination unit 7
It consists of a single character processing section 81 and a graphic processing section 9.

The image input unit 1 digitizes and inputs a document image containing a mixture of characters and figures.

The image memory 2 stores the digital data.

As shown in FIG. 2, the contour extraction unit 3 performs a logical AND operation between the original III image and an image obtained by shifting the original image by 1m elements to the right, and combines the resulting image and the original image by 1 pixel. , perform a logical AND with the image shifted to the left, and similarly add 1 to the top and bottom.
If we take the logical product with the image shifted by the pixel, we get 1 in 4 directions.
An image that has been shrunk by a pixel is completed, and the contour is extracted by performing an exclusive OR of this image and the original image.

FIG. 3 shows the original image, the contracted image, and the contour extracted image.

The chain encoding unit 4 extracts the contour extracted by the contour extraction unit 3! Perform raster scanning on the ii image, find a black pixel, replace the black pixel with a white pixel, scan 8 neighborhoods of that pixel, and if there is a black pixel, chain according to Figure 4. Encoding is performed, the pixel is replaced with a white pixel, and this process is repeated one after another until there are no black pixels in the 8-neighborhood. Furthermore, if there are a plurality of black pixels in the 8-neighborhood, position information of black pixels other than the pixel of interest is stacked.

When the main black pixel tracking is completed, the stacked pixel position information is taken out and chain encoding is repeated in the same way. Since the chain code traced and extracted from this stacked pixel information is connected to the main chain code, it is treated as a child chain code. Once all the pixel information in the stack has been extracted, the raster scan that was initially performed continues. This results in chain encoding of all images. FIG. 5 is a flowchart showing the chain encoding procedure in more detail.

First, the pixel of interest is set at the upper left of the image as the starting point of scanning (
Step 11). It is checked whether the pixel of interest is a black pixel (step 12), and if it is not a black pixel, the pixel of interest is changed according to the raster scanning order (step 13). It is determined whether the pixel of interest is the final pixel (for example, the lower right pixel of the image) (step 14), and if it is not the final pixel, the process returns to step 12, and it is determined whether the pixel of interest is a black pixel, If the pixel is not a black pixel, the pixel of interest is changed, and scanning is repeated in accordance with the raster scanning order until the pixel of interest encounters a black pixel.

When the pixel of interest is a black pixel, the pixel of interest is changed to a white pixel (step 15), and it is checked whether there are black pixels in 8 neighborhoods around the pixel of interest (step 1B).
). When there are 2 or more black pixels in the 8 neighborhood, 2 are added to the stack.
Enter the coordinates of the item or more (step 18). For the first black pixel, perform the process of the next step (19).
If there is one black pixel near the pixel of interest, it is chain encoded (step 19). That is, it is encoded using the numerical values (0, 1, 2, 3, 4 degrees, 5, 6, 7) in the corresponding direction among the directions of eight neighboring pixels as seen from the center pixel in FIG. Then, the black pixel in the vicinity is set as the next pixel of interest, and the process returns to step 15 to continue the same chain encoding process.

If there is no black pixel in the eight vicinity of the pixel of interest, the stack is checked (step 20). If there are black pixel coordinates in the stack, the coordinates are taken out in step 21), these are taken as the pixel of interest, and the process returns to step 15 to continue the same chain encoding process. If there are no black pixel coordinates in the stack, the process returns to step 12 to scan the next black pixel.

The position of the pixel of interest at this time is the coordinate before performing the tracking process for chain encoding.

When the last pixel of the scan is reached (step 14), the chain encoding process ends.

The direction change counting unit 5 counts the change in direction of each chain-coded black picture (a set of one main chain code and child chain codes connected thereto).

The black connection counting unit 6 counts the black pixels of the original image that are connected to the series of chain codes (a set of a main chain code and its child chain codes) generated by the chain encoding unit 4. For example, the contour extracted image in Fig. 3(C) is a series of chain codes (the "wo" hole, that is, 10 in Fig. 3(C)) is in contact with the outer contour, and is a series of chain codes. The black pixels of the original image connected to this chain code form the contracted image shown in FIG. 3(b). That is, counting the black pixels connected to a series of chain codes is equivalent to counting the area of the isolated image ~
I will do it. In addition, if there are multiple outlines as shown in Figure 6, that is, multiple sets of chain codes form one isolated image, those multiple sets of chain codes are newly created as a series of chain codes. , direction change counting section 5
Corresponds to the chain code being processed in . Specifically, such a chain code number is notified to the character/graphic area determining section 7.

The area determination unit 7 uses the fact that the four characters are relatively small (i.e., they are complex and the change in the direction of the outline is cumbersome, and the figure is relatively thick (i.e., they are simple and the change in the direction of the outline is small) to determine the size of each character. The number of changes in the direction of the chain code and the number of pixels of its black connected image (
For example, the relationship (area) as shown in equation (1) is taken, and characters and figures are separated at a certain threshold.

(Number of changes in direction of chain code)/(Area) Areas determined to be characters are passed to the character processing unit 8, and areas determined to be graphics are passed to the graphic processing unit 9.

The character processing unit 8 and the graphic processing unit 9 perform processing suitable for the characteristics of each image, such as character recognition for document image recognition and vectorization of graphics.

In the above embodiment, the contour extraction unit 3 contracts in four directions: top, bottom, left, and right, but it is now possible to perform contraction in eight directions, including top right, bottom right, top left, and bottom left. It can also be done.

In addition, although a contracted image is used as the contour extraction unit to extract the contour in the above embodiment, other contour extraction means,
For example, a method may be used in which the contour is obtained by performing differential processing on the image.

In addition, in the above embodiment, chain encoding is performed after contour extraction, but instead of performing contour extraction, chain encoding is performed by tracing the boundary between black pixels and white pixels. You can also do that.

(Effects of the Invention) The present invention encodes a contour line into a chain code, counts the number of changes in direction from the chain code, and counts the number of black pixels (area) connected to the chain code. Since the structure is configured to separate the character area and the figure area, large characters can be separated as characters, and small figures can be separated as figures.Moreover, the process is relatively simple compared to the conventional technology, so the processing speed can be increased. , even if characters and graphics are mixed in a complex manner (1), character areas and graphic areas can be reliably determined.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the configuration of an embodiment of the present invention. FIG. 2 is an explanatory diagram of a method for creating a contracted image. FIG. 3(a) shows an original image, FIG. 3(b) shows a contracted image obtained by the method of FIG. 2, and FIG. 3(c) shows an example of an image from which contours have been extracted. FIG. 4 is a diagram showing the direction of chain codes and the numbers representing them when chain encoding is performed. FIG. 5 is a flowchart showing the chain encoding procedure in this embodiment. FIG. 6 is a diagram showing an example in which a plurality of chain codes exist for one black connected image. 1... Image input section, 2... Image memory, 3...
- Contour extraction unit, 4... Chain encoding unit, 6... Direction change counting unit, 6... Black connection counting unit, 7... Area determination unit, 8... Character processing unit, 9. ...Graphic processing section. Figure 1 Figure 2 Figure 3 Figure 4

Claims (1)

[Scope of Claims] An image input unit that digitizes and inputs an image; an image memory that stores the image; a contour extraction unit that extracts a contour in the image; and a contour extraction unit that tracks the contour extracted by the contour extraction unit; A chain encoding unit that chain-codes this, a direction change counting unit that counts the number of changes in the direction of the chain code, a black connection counting unit that counts the black pixels connected to the series of chain codes, and a direction change counting unit that counts the number of black pixels connected to the series of chain codes. A text/graphic area determining device comprising: a region determining unit that determines a character area and a graphic area based on the number of direction changes counted by a counting unit and the number of black pixels counted by a black connection counting unit.