JPH0350691A - Character recognizing device - Google Patents

Abstract

PURPOSE:To simply recognize the character of equal connection information like 'b' or 'd' by providing a character segmenting part, a feature point detecting means, a direction detecting means for a segment and a character discriminating means. CONSTITUTION:In the case of the character 'b', an input part 1 receives a picture including a character pattern as a binary picture, and stores it in a character pattern area 23. The character segmenting part 2 segments an object character pattern to be recognized. Next, a feature extracting part 3 makes the pattern thin so as to convert it into a center line pattern. The existence of an inflection point is decided in respect of a picture element on a center line to a branch point (b) as considering a scanned and detected point (a) a beginning point. When difference between the length of the center line from the beginning point to the corresponding picture element and the shortest distance is over a threshold, a middle point on the center line is made the inflection point, and a vector direction from the beginning point to the inflection point is given by an octas code. Finally, a recognizing part 5 selects the character having a feature stored in a feature area 24 corresponding to the number and the direction code of strokes connecting an end point, the branch point and an intersection point and the direction code of the segment at the branch point from a dictionary area 28, and stores it in a character code area 25, and displays it on a display part 32, and then, 'b' and 'd' can be discriminated effectively from each other.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of Industrial Application The present invention reads printed characters, dot characters, and handwritten characters from newspapers, magazines, etc., and converts input characters (turns) into JIS
The present invention relates to a character recognition device that converts into code information such as a code.

2. Description of the Related Art In recent years, character recognition devices have come to be commonly used as character input devices for personal computers and the like. Character recognition devices for alphanumeric characters, kana characters, and symbols include, for example, "Recognition method for handwritten English, numbers, symbols, and katakana characters using a top-down method" (Hoshino, Kami, Shina, Ito, Ueda, Kushinami) have been announced.

The operation of the conventional character recognition device will be explained below.

First, the original character pattern is smoothed and thinned to obtain a core pattern. Next, rask scanning is performed to detect end points and branch points, and if the length (stroke) between the end point and the end point or branch point connected thereto is shorter than a threshold value, the stroke is removed from the skeleton pattern. When the rask is scanned again and an end point is detected, the stroke is continued from the end point as the starting point until it encounters an end point or a branch point. If the core wire is bent in the middle, give a bending point. Next, a direction code string is given to the stroke from the starting point to the end point or branch point.

Characteristics are obtained by arranging the stroke direction code sequences, end points, bending points, and branching points obtained from the above in the order of detection.

The feature is then compared with standard pattern features stored in a dictionary to determine a candidate category.

Finally, using a decision tree that uses features and thresholds to distinguish the paired categories combined from the candidate categories, one is eliminated in a tournament style, and the remaining category is taken as the recognition result.

Problems to be Solved by the Invention However, with the above conventional configuration, it is difficult to recognize characters such as "b" or rdJ because the characteristics used in classification may be the same, and it is difficult to recognize characters in detailed classification. I had no choice but to do it. Due to this low classification ability,
The problem is that the dictionary for detailed classification becomes complicated and the burden of creating the dictionary increases.

Means for Solving the Problems In order to solve the above problems, the present invention arranges stroke direction code strings, end points, bending points, and branching points in the order of detection from a core line pattern obtained by thinning an original character pattern. Recognition is performed using not only the features but also the combination of the directions of the three core lines extending from the branch point.

Effects Since the present invention extracts combinations of directions of three core lines extending from a branch point with the above-described configuration, even characters having the same connection information, such as b and d, can be easily recognized.

Embodiment FIG. 1 shows a functional block diagram of a character recognition device in an embodiment of the present invention. In FIG. 1, 1 is an image input unit composed of a scanner etc. that reads image data;
2 is a character extraction unit that extracts characters from the image read by the image input unit 1, 3 is a feature extraction unit that extracts features from the original character pattern extracted by the character extraction unit 2, and 4 is a character that corresponds to a combination of features. 5 is a character recognition unit that searches the dictionary 4 for a character code corresponding to the feature extracted by the feature extraction unit 3; 6 is a display unit.

FIG. 2 is a block diagram showing the configuration of the character recognition device of this embodiment. Here, 21 outputs the character pattern read by a scanner that reads the character pattern as bit data. Reference numeral 22 denotes a character pattern area 23 in RAM that stores bit data from the scanner 21, and a characteristic area 2 obtained by analyzing the character pattern within this character pattern area 23.
4. Character code area 25 determined from the features stored in the feature area 24 and register area 26 used in processing
have. Reference numeral 27 is a ROM, and has a dictionary area 28 that stores features and character codes corresponding thereto, and a program storage area 29 that stores a control program according to the flowchart shown in FIG. 30 is a processing circuit that performs processing according to a control program stored in the program storage area 29. 31 is a keyboard for inputting data, and 32 is a display section for displaying characters corresponding to the character codes stored in the character code area 25.

The character recognition device of this embodiment configured as described above will be explained according to the flowchart shown in FIG. In step S1, the character pattern is thinned to obtain a core pattern. In step S2, the skeleton pattern is scanned from the upper left to the lower right, and when the pixel of interest in scanning (scanning pixel) overlaps the skeleton pattern, scanning is stopped, and step S3
Proceed to. In step S3, it is determined whether the scanning pixels overlapping the skeleton pattern are end points. Here, among the eight pixels where the scanning pixel is a black pixel and is adjacent to the black pixel, 1.3.
A point where four pixels are black is called an end point, a branch point, and an intersection point, respectively. If the scanned pixel is not an end point, the process jumps to step S11, and if it is an end point, the process jumps to step S4. In step sll, it is determined whether or not the scanning of the skeleton pattern has been completed. If it has not been completed, the process returns to step s2, but if it has been completed, the feature extraction process is ended. Next, in step S4, line segment tracing is performed.

That is, the end point is the starting point, and the process proceeds by one pixel along the skeleton extending from the starting point. The point advanced by one pixel is set as the point of interest. Next, the process proceeds to step S5, where the shortest distance from the starting point to the point of interest is compared with the length of the core line to determine whether it is a bending point.

If the difference between both is greater than or equal to the threshold, the process goes to step s12 to give an inflection point, and if it is less than the threshold, the process goes to step S6. In step s12, the number and position of bending points are stored in the register area 26, and the process proceeds to step S6. In S6, it is determined whether the point of interest is an end point. If it is an end point, the process advances to step s13; if it is not an end point, the process advances to step s7. Step S7
Then, it is determined whether or not the point of interest is an intersection. If it is an intersection, the process proceeds to step slo; otherwise, the process proceeds to step S8. In step s13, the length of the skeleton from the starting point to the end point is calculated, and if the length is less than the threshold value, it is regarded as noise called "whiskers", and the process returns to step S2 without extracting any features and resumes scanning, but the length is longer than the threshold value. In this case, the process advances to step s18.

In step slO, the length of the skeleton from the starting point to the intersection point is calculated, and if the length is less than the threshold, the process returns to step S2,
In other cases, the process proceeds to step s14. Step S8
Then, it is determined whether the point of interest is a branching point or not. If it is a branch point, the process advances to step S9; otherwise, the process returns to step $2. In step s9, the length of the skeleton from the starting point to the branching point is calculated, and if the length is less than the threshold value, it is regarded as a whisker, and the process returns to step s2 without extracting any features and the scanning is restarted. If so, the process proceeds to step s15. In step s15, a direction code is assigned to the stroke from the starting point to the branch point. Next, in step s16, a line segment is traced by a fixed number of pixels in three directions with the branch point as the starting point.

That is, the branch point is the starting point, and three extraction points are obtained by moving forward by a certain number of pixels along the skeleton lines in three directions extending from the starting point. Next, in step s17, the directions from the branch point to the three extraction points are calculated, a direction code is assigned, and step s
Return to 2. Step S14. In step s18, direction codes are assigned to the strokes from the starting point to the intersection point and from the starting point to the end point, respectively, and then the process returns to step s2.

The operation will be explained below using the recognition example character rbJ as an example.

First, the image input unit 1 inputs an image including a character pattern to be recognized as a binary image and stores it in the character pattern area 23 .

Next, the character cutting section 2 cuts out a character pattern to be recognized from the input image stored in the character pattern area 23. The pixel values of the character part and the background part of the input character pattern are each 1.0. Furthermore, the pixels forming the character part and the background part are called black pixels and white pixels, respectively.

Next, the feature extraction unit 3 converts the character pattern into a skeleton pattern by thinning. The skeleton pattern of the recognition example character pattern rbJ is shown in FIG. 4(a). Next, the skeleton pattern is scanned, the detected point a is used as the starting point, and line segment tracing is performed, and for each pixel on each skeleton until the branch point is encountered,
Determine whether an inflection point exists. In other words, when the difference between the length of the skeleton and the shortest distance from the starting point to the relevant pixel is greater than the threshold, the midpoint on the skeleton from the starting point to the relevant pixel is set as the bending point, and the direction of the vector from the starting point to the bending point is set in 8 directions. Grant with code. In the case of point a-s, there is no bending point because the difference is less than the threshold value. Also, since the length between point a and point b is greater than the threshold value, it is not a beard, so as shown in Figure 4 (b),
Give the direction of the line segment from point to point b using a direction code. If it is less than the threshold value, such as point 0-d, it is removed. Points obtained by tracing a line segment by a certain number of pixels in three directions starting from point b are designated as e, f, and g (FIG. 4(C)).

The direction from point b to point e and point 1 to point 1g is calculated, and the direction code 1, 3.7 is obtained. The direction code is shown in FIG. 4(d). From the above, it can be seen that there is a stroke connecting the end point and the branch point, that the stroke direction has a direction code of 7, and that the line segment direction code at the branch point is 137.

Finally, the character recognition unit 9 corresponds to the number and direction code of strokes connecting end points and end points, end points and branching points, and end points and intersections stored in the feature area 24, and the direction code of the line segment at the branching point. Characters having standard pattern characteristics are searched from the dictionary area 28, the corresponding character code is stored in the character code area 25, and displayed on the display section 32.

As described above, according to this embodiment, since the directions of the three core lines extending from the branch point are featured, even characters that could not be distinguished using conventional features alone can be effectively recognized.

Effects of the Invention The present invention is characterized by the direction of a plurality of core lines extending from a branch point, thereby realizing an excellent character recognition device that can effectively recognize characters that could not be distinguished using conventional characteristics alone. It is.

[Brief explanation of drawings]

FIG. 1 is a functional block diagram of a character recognition device according to an embodiment of the present invention, FIG. 2 is a block diagram, FIG. 3 is a flowchart, and FIG. 4 is an explanatory diagram of the present invention. 1... Image input section, 2... Character extraction section, 3... Feature extraction section, 4... Dictionary, 5... Recognition Section, 6...display section.

Claims (1)

[Claims] a character cutting section that cuts out a character pattern to be recognized from an input image; a feature point detection means that detects an end point, a branch point, or an intersection point from the character pattern cut out by the character cutting section; and a certain detected branch point. A character pattern to be recognized is recognized by a combination of a detection means for detecting the direction of each of a plurality of line segments from to another end point, branch point or intersection, and the direction of the line segment at a certain branch point detected by the detection means. What is claimed is: 1. A character recognition device comprising character recognition means.