JPH01187684A - Character recognizing device - Google Patents

Abstract

PURPOSE:To correctly recognize even a character in which a crush is apt to occur by classifying a character based on the direction code frequency distribution value of the outermost point which is not influenced by the crush of the character. CONSTITUTION:In an outermost direction code detecting means 21, a black picture element to meet first by scanning from the upper, the lower, the right and the left of a character frame 8a toward a character part in directions perpendicular to the character frame 8a is detected as the outermost point strong against the crush and the direction code 18 of the outermost point is determined. After that, the outermost direction code frequency distribution value obtained by counting the appearance frequency the direction code 18 of the outermost point to meet by scanning from points on respective divided character frames into which the character frame 8a is halved lengthwise and breadthwise is matched with a standard feature value by a matching means 24. Thus, even a character pattern in which the crush occurs can be correctly recognized.

Description

[Detailed Description of the Invention] [Industrial Application Field] This invention relates to a character recognition device that reads many character types.
In particular, the present invention relates to a character recognition device that recognizes kanji characters, which are prone to font variations, with high accuracy.

[Prior Art] As a means of inputting Japanese information into a computer, there are increasing expectations for a character recognition device that can directly input a single handwritten kanji character and recognize characters. As a character recognition method for such a character recognition device, a method has been proposed that uses a structural analysis method to find character features and recognizes the characters using a one-superposition method.

In order to read characters with high accuracy, it is necessary to classify characters using features that are stable even when changes such as differences in font or thickness of character lines occur.

For example, Figure 4 shows the Institute of Electronics and Communication Engineers study group material PRL82-
FIG. 2 is a block diagram showing the configuration of a conventional character recognition device shown in "Study of Handwritten Kanji Recognition Using Stroke Matching Method Using 3Or Global Features". The stroke matching method that uses global features is a method in which the contour points of a binary pattern are given four types of orientation: horizontal, vertical, upward-to-the-right, and upward-to-the-left, and the direction code distribution feature is used to calculate the number of occurrences. Extract and thin the binary pattern to find end points, branch points,
This is a method of extracting feature points such as refraction points, and then extracting the feature point distribution feature by calculating the number of feature points. In FIG. 4, 1 is a sheet of paper with characters written on it, 2 is a scanning means for optically scanning the characters written on the sheet 1 and photoelectrically converting the characters, 3
4 is a contour direction code detection means for detecting the direction code of the contour points of the photoelectrically converted character pattern, and 4 is a contour direction code frequency distribution value indicating the number of direction codes of the contour points detected by the contour direction code detection means 3. 5 is a standard feature value storage means that stores contour direction code frequency distribution values representative of characters to be recognized as standard feature values; 6 is a contour direction code frequency distribution value calculation means 4;
7 is a matching means for matching the contour direction code frequency distribution value calculated by the standard feature value stored in the standard feature value storage means 5 and determining the degree of matching between the two; ), and selects the character with the smallest difference value among all recognition target characters as a recognition result, and outputs it to an external device.

5(A) to (D>) are operation explanatory diagrams for explaining the operation of the conventional device shown in FIG. 4. 8 shown in FIG. Example, Figure 5 (B
9 shown in ) is an example of assigning direction codes to the contour points of character pattern 8, and 10 shown in FIG. 5(C) is character pattern 8.
This is an example in which the area is divided into four. 11 shown in Figure 5(D)
is an example of the direction code frequency distribution value of contour points in each divided area. H, R, V, and L in figure 1 indicate that the corresponding contour points of character pattern 8 are horizontal, upward-rightward, vertically, and upward-leftward, respectively. .

Figure 6 is an example of a 3x3 mask pattern for determining whether the contour points are horizontal, upward to the right, vertical, or upward to the left, and 12 in figure 9 is the black pixel of interest. . 7(A) to (C) are examples of the processing results of the conventional apparatus shown in FIG. 4 for characters with blurring, and 13 in FIG. 1 is an example of a character pattern with blurring.
14 shows an example in which direction codes are assigned to contour points of a character pattern, and 15 shows an example of direction code frequency distribution values of contour points.

Next, the operation of the conventional device shown in FIG. 4 will be explained using FIGS. 5 and 6. First, the character pattern 8 obtained by photoelectrically converting the characters on the paper 1 by the scanning means 2 is transferred to the contour direction code detecting means 3. Next, while mask-scanning the 9-character pattern 8 from the upper left to the lower right, the contour direction code detecting means 3 detects all the contour points of the character pattern 8 using the 3×3 mask pattern shown in the example of FIG. It is detected which direction code each contour point is assigned to among the four directions, and the direction code 9 for all the contour points of the character pattern is determined as shown in the example of FIG. 5(B). The direction code 9 of the contour point obtained by the contour direction code detection means 3 is sent to the contour direction code frequency distribution value calculation means 4. Then, the contour direction code frequency distribution value calculation means 4 divides the character pattern into 2×2 areas as shown in the example of FIG. Find the number of direction codes for V and left-upward L, and calculate the contour direction code frequency distribution value F(fl,
f2. −・−, fM)=(20,4,8,2,−−,
14, 7, 3, 8). Next, in the matching means 6, the contour direction code frequency distribution value F(fl, f2...,
fl4) and the contour direction code frequency distribution value Tk (jl
, t2. . Here, as the degree of difference Dk, the absolute value distance shown in equation (1) is used.

Finally, the classification means 7 selects the character having the smallest difference value among all the characters to be recognized as the recognition result and outputs it to an external device.

[Problems to be Solved by the Invention] Conventional character recognition devices are configured to recognize characters based on the number of direction codes of all outline points of a character pattern as described above, so nine character lines are collapsed. The problem is that the contour direction code frequency distribution value extracted from a character pattern in which the internal contour shape of the character has been deformed is significantly different from the contour direction code frequency distribution value extracted from a character pattern without collapse etc. was there. for example.

Character pattern 13 with collapse as shown in FIG. 7(A)
In this case, the contour points inside the character no longer exist.

In a conventional character recognition device, a direction code 14 is assigned as shown in FIG. 7(B). As a result, the contour direction code frequency distribution value F'(r
i, f2', ..., fM) = (6, 1, 8, 0,
..., 8, 4, 3.8>.

Contour direction code frequency distribution value F (fl, f2..., f
M) = (20, 4, 8, 2,..., 14, 7, 3.
8> is significantly different.

As described above, the conventional device uses the frequency of the direction code assigned to all outline points of a character as a feature value for recognition, which has caused the problem that character patterns with blurring or the like cannot be correctly recognized.

The present invention has been made to solve the above-mentioned problems, and it is an object of the present invention to provide a character recognition device that can correctly recognize even characters that are distorted or otherwise distorted.

[Means for Solving the Problems] A character recognition device according to the present invention scans from a point on a character frame 8a circumscribing a single character pattern 8 toward an internal character portion in a direction perpendicular to the character frame 8a. outermost direction code detection means 21 for detecting the direction code 18 of the outermost point indicated by the black pixel that encounters the outermost direction code detection means 2;
outermost direction code frequency distribution value calculation means 22 that calculates the frequency distribution value of the outermost direction code based on the information of the direction code 18 of the outermost point detected in step 1; The standard feature value storage means 23 stores distribution values as standard feature values, and the outermost direction code frequency distribution value calculated by the outermost direction code frequency distribution value calculation means 22 and the standard feature stored in the standard feature value storage means 23 The apparatus is characterized in that it has a matching means 24 that matches the values and determines the degree of matching between the two, and recognizes characters written on paper 1 or the like based on the degree of matching.

[Operation] The outermost direction code detection means Z/ scans from the upper, lower, left, and right character frames 8a toward the inner character portion in a direction perpendicular to the character frame 8a, and sets the first black pixel encountered as the outermost point;
The direction code 18 at the outermost point is detected. The outermost direction code frequency distribution value calculating means 22 calculates the frequency distribution value of the outermost direction code based on the detected information. The matching means 24 matches the outermost code frequency distribution value and the standard feature value to determine the degree of matching between the two. Therefore, paper 1
The character written in is recognized as the character having the smallest value of the degree of consistency (degree of difference).

Embodiment of the Invention FIG. 1 is a block diagram showing the configuration of a character recognition device according to an embodiment of the invention. In fig.

1 is a sheet of paper on which characters are written; 2 is a scanning means that optically scans the characters written on the sheet 1 and converts them photoelectrically; 21 is a scanning device that scans the characters written on the sheet 1 from a point on the character frame circumscribing the character pattern toward the internal character area; outermost direction code detection means for scanning in a direction orthogonal to the character frame and detecting the direction code of the outermost point indicated by the first black pixel encountered.

22 is outermost direction code frequency distribution value calculation means for calculating the frequency distribution value of the outermost direction code based on the information of the direction code of the outermost point detected by the outermost direction code detection means 21; 23 is a means for calculating the frequency distribution value of the outermost direction code; The standard feature value storage means 124 stores representative outermost direction code frequency distribution values as standard feature values, and the outermost direction code frequency distribution values calculated by the outermost direction code frequency distribution value calculation means 22 and the standard feature value storage means 23 A matching means 25 matches the standard feature values stored in , and calculates the degree of matching between the two, and 25 classifies the characters based on the matching degree (difference degree) obtained by the matching means 24 and calculates the degree of dissimilarity of all characters to be recognized. This classification means selects the character with the smallest value as a recognition result and outputs it to an external device.

Figures 2 (A) to (C) are operation explanatory diagrams for explaining the operation of this embodiment. 8 in Figure 1 is a character pattern written on paper 1, and 8a is a character circumscribing character pattern 8. Frame 18 shows an example of the direction code assigned to the outermost point of character pattern 8, and 19 shows an example of the outermost direction code frequency distribution value obtained by dividing the hand-knit horizontal character frame. Note that H, R, V, and L in Figure 2 are horizontal and upward to the right, respectively, as in the example in Figure 5.

Indicates that it is vertical, rising to the left. Also, Figure 3 (A)
, (B) are diagrams illustrating an example of the processing results of this embodiment for a character pattern in which distortion has occurred.

13 in the figure is an example of a character pattern in which collapse has occurred.

20 shows an example of the outermost direction code frequency distribution value.

Next, the operation of the embodiment shown in FIG. 1 will be explained using FIGS. 2 and 3. In this embodiment, the outermost direction code detection means 21 detects a direction perpendicular to the character frame 8a from each point on the character frame 8a circumscribing the character pattern 8 photoelectrically converted by the scanning means 2 toward the inner character part. The first black pixel encountered during scanning is determined to be the outermost point, and a direction code 18 is assigned to this outermost point.7 The direction code 18 is detected.

Further, the outermost direction code frequency distribution value calculation means 22 is as follows.

The outermost direction code frequency distribution is obtained by scanning from each point on each divided character frame obtained by dividing the character frame 8a vertically and horizontally into N equal parts, and finding the frequency of the direction code of the outermost point encountered for each divided character frame area. The value shall be calculated.

First, the character pattern 8 obtained by photoelectrically converting the characters on the paper 1 by the scanning means 2 is transferred to the outermost direction code detection means 21. Next, in the outermost direction code detection means 21, 1
Scanning is performed in the vertical direction from each point on the upper end of the character frame 8a circumscribing the character pattern 8 toward the character part, and the first black pixel encountered is detected as the outermost point.
The direction code of this black pixel is determined using a ×3 mask pattern. Thereafter, in the same manner, scanning is performed sequentially from the lower end, left end, and right end of the one-character frame 8a toward the inner character portion.

As shown in the example of FIG. 2(B), direction codes 18 in four directions, ie, horizontal H1, upward-rightward R1, vertical V, and upward-leftward L, are assigned to all the outermost points of the character pattern 8. Information on the direction code 18 of the outermost point detected by the outermost direction code detection means 21 is sent to the outermost direction code frequency distribution value calculation means 22. Then, the outermost direction code frequency distribution value calculating means 22 reduces and horizontally divides the character pattern into two equal parts, and scans from a point within each divided character frame area, as shown in the example of FIG. 2(C). Find the frequency of the direction code of the outermost point encountered by
, 8, sail..., 3, 1, 0°5). However, N=32.

FIG. 3 shows the outermost direction code frequency distribution value G (gl,
g2',..., gN)・(0,1,8,0,...,
3.2, 0.4) is calculated. However, N=3
It is 2. Since the outermost direction code frequency distribution value uses information on the outermost point of the character pattern that is less susceptible to collapse, the second point without the character part shown in the example of Figure 3 (A) where collapse occurs Outermost direction code frequency distribution value G(gl, g2...,
gN)=<0.1, 8, 0. ..., 3, 1, 0.5>
and take almost the same value.

In other words, the degree of difference indicating the degree of difference between the two is expressed as Equation 1%. By extracting the outermost direction code frequency distribution value, stable feature values that are not easily affected by the collapse of one character can be obtained.

Outermost direction code frequency distribution value G obtained as above
(gl,, g2..., gN) is the matching means 24
sent to. Then, in the matching means 24, the outermost direction code frequency distribution value G (gl, g2..., gN
) and the outermost direction code frequency distribution value Pk (pi.

p2. . . , pN> in the same manner as in the conventional apparatus, and the degree of matching (difference Dk>) between the two is determined.

Finally, the classification means 25 selects the character having the smallest value among all the recognition target characters as a recognition result and outputs it to an external device.

In addition, in the above embodiment, an example was explained in which the direction code frequency distribution of the outermost point for the vertical and horizontal character frame regions divided into two equal parts is obtained, but the present invention is not limited to this.
2) The direction code frequency distribution of the outermost point for the vertical and horizontal character frame areas may be determined.

According to the above embodiment, the outermost direction code detecting means 21 scans from the upper, lower, left, and right character frames 8a toward the inner character part in a direction perpendicular to the character frame 8a, and selects the first black pixel encountered that is the most resistant to crushing. After detecting it as an outer point and determining the direction code 18 of this outermost point, the direction code of the outermost point that is encountered by scanning from a point on each divided character frame obtained by dividing the 9 character frame 8a vertically and horizontally into two is determined. By matching the outermost direction code frequency distribution value obtained by counting the frequency of appearance for each divided character frame region with the standard feature value, even a character pattern 13 with blurring can be accurately recognized.

[Effects of the Invention] As described above, according to the present invention, the first black pixel encountered when scanning from a point on a character frame circumscribing a single character pattern toward an internal character part in a direction perpendicular to the character frame indicates The direction code of the outermost point is detected, the outermost direction code frequency distribution value is calculated based on the information of this direction code, and the outermost direction code frequency distribution value is matched with the standard feature value to determine the degree of consistency between the two. Since this configuration allows characters to be classified based on the direction code frequency distribution value of the outermost point, which is not affected by the distortion of a single character, even characters that are likely to be distorted can be accurately recognized. This has the effect of improving recognition accuracy.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the configuration of a character recognition device according to an embodiment of the present invention, and FIGS. 2 (A), (B), and (C) are operation explanatory diagrams for explaining the operation of this embodiment. . 3(A) and 3(B) are diagrams showing an example of the processing results of this embodiment for characters with blurring, FIG. 4 is a block diagram showing the configuration of a conventional character recognition device, and FIG. ), (
B), (C), and (D) are operation explanatory diagrams for explaining the operation of this conventional example, and FIG. 6 is an example of a mask pattern for determining the direction of the outermost point in the embodiment and the conventional example. The figures shown in FIGS. 7(A), 7(B), and 7(C) are diagrams showing examples of processing results of the conventional apparatus for characters with blurring. 1...Paper, 8...Character pattern,
8a...Character frame, 18...Outermost point direction code 721...Outermost direction code detection means, 22...Outermost direction code frequency distribution Value calculation means, 23...Standard feature value storage means. 24... Coordination means.

Claims (1)

[Claims] Outermost direction code detection means for detecting the direction code of the outermost point indicated by the first black pixel encountered by scanning from a point on the character frame circumscribing the character pattern toward the inner character part in a direction perpendicular to the character frame. and outermost direction code frequency distribution value calculation means for calculating the frequency distribution value of the outermost direction code based on the information of the direction code of the outermost point detected by the outermost direction code detection means, and standard feature value storage means for storing the outermost direction code frequency distribution value as a standard feature value; and the outermost direction code frequency distribution value calculated by the outermost direction code frequency distribution value calculation means and the standard feature value storage means. What is claimed is: 1. A character recognition device comprising: matching means for matching stored standard feature values to determine a degree of matching between the two, and recognizing characters written on paper or the like based on the degree of matching.