JPS62298888A - Input decision system for print character - Google Patents

Abstract

PURPOSE:To decide the inclination of an input character at high speed with high accuracy by calculating the input angle of a character string from the peaks of direction histograms of edges found by taking the first derivative of a character print area. CONSTITUTION:A direction histogram calculating means 1 takes the first derivative of the character print area of an input image to find direction histograms of respective edges. A character string input angle calculating means 2 calculates the input angle of the character string from the peak values of the direction histograms calculated by the direction histogram calculating means 1. An input angle deciding means 3 compares the character input angle with a preset permissible range angle to decide whether or not the input angle is within the permissible range. Then when the input angle of the character string is within the permissible range, the segmentation processing and then recognition processing of characters are performed and when not, on the other hand, prompting for reinput or the rotation of the input image by the calculated character string input angle is performed to perform the character segmentation and recognition processing.

Description

[Detailed Description of the Invention] 3. Detailed Description of the Invention [Summary] By obtaining an edge direction histogram from the character printing area in the image data of an input document in which printed characters are written, it is possible to determine the printed character string. The direction of the input character string is detected, and it is determined whether the direction of the input character string is within the allowable range for subsequent processing.

[Industrial Field of Application] The present invention relates to pattern recognition, and particularly to input in printed character recognition.

The scope of character recognition has been expanding from printed characters to handwritten characters, but with printed characters, the angle of the character string relative to the sensor that reads the character image poses a problem. This is because, in the subsequent processing, character extraction is performed based on projection, and the inclination of characters causes a reduction in recognition rate in a printed character recognition system based on pattern matching.

Therefore, it is necessary to determine whether the input angle of the character string is within the allowable range of the character extraction method and recognition method.

[Prior Art] Conventionally, there have been two methods for determining whether or not a printed character has been input at an angle within a permissible range.

■A method of calculating the projection while changing the calculation direction and detecting the angle at which the character string is printed based on the direction in which the variance of the projection is maximum. That is, if the direction of projection matches the direction of the character string, spaces between lines will appear in the projection data, and the variance will be maximized.

■A method of detecting the angle of a character string by calculating the projections for each small vertical area, finding the correspondence between them, and applying a straight line.

As shown in Figure 5 (al), if the direction of the printed character string matches the direction in which the projection is sought, the space between the lines will be visible, but if the direction of the printed character string does not match the direction in which the projection is sought, as shown in the mountain, the space between the lines will be visible. I can't see the sky. Therefore, we divide the image perpendicularly to the direction in which we want the projection and find the projection (as shown in C1,
I see a blank space.

Looking at the projection data in (C), it can be seen that the projection is shifted in a direction perpendicular to the direction in which the projection is sought. Find the direction of the string from this amount of movement.

I can do it. In the case of (dl in the figure), the rotation angle of the character string is 2 in the x direction and 4 in the y direction, so 2
It can be calculated as 2.5°.

[Problems to be Solved by the Invention] The problem with the above conventional method is that in the method (2), the calculation direction of the projection where the variance is maximum is not known at all, so the direction is changed with a certain precision and the calculation is repeated several times. In the method (2), the actual projections cannot be obtained as neatly as shown in Figure 4, so if the correspondence between the projections can be clearly obtained, is not limited. Also, it takes time to divide and calculate.

There is a problem.

The present invention aims to provide a novel input determination method that solves these conventional problems.

Means for Solving Problem E] FIG. 1 shows a principle block diagram of the printing character input determination method of the present invention.

In the figure, l denotes a direction histogram calculation means, which performs first differentiation on the character printing area of the input image to obtain the direction histogram of each edge.

Reference numeral 2 denotes a character string input angle calculation means, which calculates a character string input angle from the peak value of the direction histogram calculated by the direction histogram calculation means 1.

Reference numeral 3 denotes an input angle determining means, which compares the character input angle calculated by the character string input angle calculating means 2 with a preset allowable range angle to determine whether or not it falls within the permissible range.

If it is determined that the human input angle of the character string is within the allowable range, character extraction processing and then recognition processing are performed.

If the input angle of the string is determined to be outside the allowable range,
It notifies you of this and prompts you to re-enter it, or it rotates the input image by the calculated character string input angle to perform character extraction and recognition processing.

[Operation] Japanese characters on printed matter are mainly composed of vertical and horizontal directions, and in addition, line segments inclined at 45 degrees from these directions.

That is, if we look at the distribution of angles of line segments in the character printing area of printed matter, it is 0 degrees, as shown in FIG.

It has very high peaks at four angles: 90 degrees, 180 degrees, and 27 (j degrees), and slightly high peaks at four angles in between.

In the present invention, in order to find this angle, the character printing area is
By scanning once with the −th order differential filter, it is possible to obtain a direction histogram of the edges of line segments constituting a character, and to calculate the angle from the peak to the reading sensor of the printed matter.

[Example] The present invention will be described in more detail below with reference to Examples shown in FIGS. 3 and 4.

FIG. 3 is a block diagram of one embodiment of the present invention.

In the figure, reference numeral 4 denotes an input device that optically scans an input printed document, photoelectrically converts reflected light, and binarizes the obtained signal.

Reference numeral 5 denotes an image memory, which stores the image data obtained by converting the input device 4 data into binary values.

6 is an area separation circuit that reads out the contents of the image memory 5 and identifies the character printing area, figure/table area, and photo area;
To separate.

Reference numeral 7 denotes a part of the image memory 5, which stores coordinate data of each area separated by the area separation circuit 6.

Reference numeral 1 denotes a direction histogram calculation circuit, which calculates the direction histogram from the image memory 5 based on the area coordinate data stored in the image memory 7.
Regarding the character printing area of the input image stored in -
Perform the next differentiation to obtain the edge direction histogram for each edge. A memory 8 stores the direction histogram data calculated by the direction histogram calculation circuit 1.

2 is a character string input angle calculation circuit which finds the peak of the direction histogram stored in the memory 8 and calculates the smallest difference between that angle and any one of 0', 90', 180', and 270'. Let be the input angle of the string.

A memory 9 stores the character string input angle calculated by the character string input angle calculation circuit 2.

Reference numeral 10 denotes a memory, which stores the smaller of the character cutout portion and the permissible range of the recognition portion.

3 is a manual angle determination circuit that compares the character string input angle stored in the memory 9 with the permissible range angle stored in the memory 10, and determines whether the input was made at an angle exceeding the permissible range. judge.

FIG. 4 is a diagram showing a direction histogram calculation method using a partial differential filter in the direction histogram calculation circuit l.

For example, as shown in the figure, the filter has a size of 9 pixels (3×3), and the filter (a) and filter (bl) are used to scan the same part of the image data.

(When scanning an image as shown in C1 in the horizontal direction,
The total value of the product of each pixel of the image and the value of each element of the filter is
In the edge, the filter (al is "-33,"
In filter (b), it becomes "0".

When scanning an image as shown in (d), in the process, the filter ta) becomes "θ", and the filter (
3" are continuous in bl.

In this way, the filter (al detects the change in the y direction, and the filter mountain) detects the change in the horizontal direction.

Therefore, when scanning a 45 degree edge as shown in tel, the value is "-2" for filter (a) and 2" for filter (a).

When the same location is scanned by the filters (al and filter 01), the values obtained from both are shown in (f) in Figure 4.
As shown in , the X component and the y component in the normal direction of the edge are shown, and therefore, the direction of the cut and the cut can be determined from the ratio of these values.

If the entire character printing area is scanned once using the filters ta> and To) and a histogram is created for the edge direction determined from the ratio of the values obtained at that time, a direction histogram similar to that shown in FIG. 2 can be obtained.

[Effects of the Invention] As explained above, according to the present invention, the inclination of input characters can be determined at high speed and with high accuracy, and the effect of contributing to improving the performance of printed character recognition devices is extremely large. .

[Brief explanation of the drawing]

Fig. 1 is a block diagram of the principle of the present invention, Fig. 2 is a diagram showing the angular distribution of printed character line segments, Fig. 3 is a block diagram of an embodiment of the present invention, and Fig. 4 explains the direction histogram calculation method. FIG. 5 is a diagram showing a conventional rotation angle detection method. In the drawings, l is a direction histogram calculation means (circuit), 2 is a character string manual angle calculation means (circuit), and 3 is a manual angle determination means (circuit).
, 4 is an input device, 5.7 is an image memory, 6 is a region separation circuit, 8 is a direction histogram storage memory, 9 is a character string input angle storage memory, and 10 is a tolerance range angle storage memory. Block diagram of the principle of the present invention Figure 1 Figure 2 Block diagram of an embodiment of the present invention Figure 3 (d) (e
) (fl Figure 4 explaining the direction histogram calculation method

Claims (1)

[Scope of Claims] In a character recognition device that separates and recognizes characters one by one from an image of an input document in which printed characters are written, a direction in which a direction histogram of each edge is obtained by applying first differentiation to a character printing area of the input image is provided. Histogram calculation means (1
), a character string input angle calculation means (2) that calculates a character string input angle from the peak of the direction histogram calculated by the direction histogram calculation means (1), and a character input angle calculated by the character string input angle calculation means (2). The input angle determining means (3) compares the angle with a preset allowable angle to determine whether or not it is within the allowable range, and is configured to determine whether the character string of the input document is appropriate for subsequent processing. A printing character input determination method characterized by the following.