JPS5822479A - Character recognition device - Google Patents

Abstract

PURPOSE:To recognize handwritten characters with high precision by extracting straight segments from an input character pattern, and performing the recognition by using middle-point patterns of those strokes. CONSTITUTION:The picture signal of an input character on a slip 11 obtained through the scanning of a scanning means 12 is used to make the input character pattern into thin lines by a preprocessing means 13. Then, a stroke extracting means 14 finds feature points of terminal points, branch points, inflection points, etc., and then finds segments connecting those points. Further, the means 14 checks the direction of each segment at the feature points except the terminal points to concatenate the segments, thus extracting the strokes. Then, a stroke middle-point pattern from the center position of the stroke is generated and sent to a determining means 15. The means 15 finds the similarity of this stroke middle-point pattern to that of a previously stored reference character to decide on what is the input character.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a character recognition device for recognizing characters with many straight lines, and more particularly to a character recognition device for recognizing handwritten Chinese characters.

Conventionally, a pattern matching method has been used in devices that recognize Chinese characters, particularly devices that recognize printed Chinese characters. This method was effective for characters with a constant shape, such as printed kanji. However, as shown in FIG. 1, if the input character pattern 3 is even slightly tilted with respect to the reference character pattern 2 within the entry frame 1, the degree of similarity between the two becomes small. Furthermore, as shown in FIG. 2, when the input character pattern 5 has a different line width from the reference character pattern 4, the degree of similarity between the two also decreases.

Therefore, the pattern matching method has the disadvantage that it is difficult to obtain a high recognition rate for characters such as handwritten Chinese characters in which each stroke is tilted with respect to the reference pattern or the line width is not constant.

In order to eliminate these drawbacks, this invention is characterized by extracting straight line segments (hereinafter referred to as strokes) from the input character pattern and performing recognition using the midpoint pattern of these strokes. The purpose is to provide a character recognition device with high reading accuracy. Hereinafter, the present invention will be explained in detail using the drawings.

FIG. 3 is a block diagram showing an embodiment of the apparatus of the present invention. First, the input characters on the form 11 are scanned by the scanning means 12, and the input character pattern is thinned by the pre-processing means 13 using the obtained image signal of the input characters.
Next, the strokes are extracted from the thinned character pattern by the p-ri extraction means 14, and a stroke midpoint pattern is created from the center position and sent to the determination means 15.

Determinant! In R15, the above-mentioned stalling midpoint pattern,
The degree of similarity with a pre-stored standard character midpoint pattern is determined to determine what the input character is.

FIG. 4 shows an example in which the kanji character ``1'' is thinned by the preprocessing means 13, and the character portion 16 of the thinned input character pattern is indicated by @1''.

Figure 5 shows character part 1 of the thinned input character pattern above.
From step 6 to step 9 above, feature points such as end points, branch points, and refraction points are obtained by the extraction means 14, and a line segment connecting the feature points (
This figure shows the result of extracting feature points 11. .

17,, 1 F,, 17. , -... are indicated by one line ", and each segment 18 is indicated by @A" to @N". For example, a segment connected to the feature point 11 is indicated by s "A".
Segment 19 indicated by "@B", segment 2o indicated by "@c", segment 21 indicated by "@c", segment 22 indicated by "@D".

Furthermore, the stroke extracting means 14 extracts one stroke by examining the direction of each segment at feature points excluding end points and combining the segments. Strike
- Extraction of ri is carried out as follows.

First, a pair of segments connected to a certain feature point PI (
For K, the first segment has 411 image points PI.
11. Set the end point to PJ and the other end (terminal end) to PJ. → − Find the direction vector Pj P of the first segment,
For the second segment) K, the starting point is the feature point Pi,
A direction vector PIPk of the second segment with the other end (terminus) as Pk is determined. IIIE at feature points P and K
When the angle formed by the direction petator of the segment I and the second segment is O, the degree of coincidence of the directions of the above two segments is defined by CSO#, and a match is found among all segments connected to the feature point P. A pair of segments with the highest degree and a predetermined threshold value or more are combined to form a stalemate. Further, the process of determining the above-mentioned strike points is repeated for segments other than the above-mentioned segment pairs.

FIG. 6 is a diagram showing strokes determined by the stroke extraction means 14 using the above method. That is, segments 19 and 22 in FIG. 5 are combined to extract the string ρ-23 indicated by @111+,
and 21 are combined to extract a stroke 24 indicated by 12''. Each extracted stroke is 11' to mB1.
It is indicated by m.

Next, the stroke midpoint is determined from each of these strokes. The coordinates of the stroke midpoint are the coordinates of the start point of the stroke (X□Y,) and the coordinates of the end point (X,).

yt ), then ((Xs + Xt )/2−
(Yt +Yt)/2).

FIG. 7 is a diagram showing a stroke midpoint pattern 25 of the input character pattern extracted by the above processing, and the stroke midpoints are numbered in correspondence with the stroke numbers in FIG. 6.

FIG. 8 is a diagram showing the stroke midpoint pattern 34 of the reference character of the Kanji T obtained by the same process. Determination means 15
, stroke midpoint pattern 2 of the input character above
5, the stroke midpoint pattern 34 of the reference character stored in the determining means ISK, and bi, the similarity is determined to determine what the input character is.

Specifically, first, each stroke midpoint of the reference character is 35 to 4.
2 is associated with the closest point among the stroke midpoints 26 to 33 of the input character.

In this example, stroke midpoints 26 to 33 are associated with stroke midpoints 3s to 42 of the reference character.

Next, the distance between the two correlated points is added, and the reciprocal of the distance is taken as the degree of similarity of the input character to the reference character. Finally, the reference character with the largest l114Ji degree is determined as the recognized character. Since recognition is performed using the stable stroke midpoint, higher recognition accuracy can be obtained than in the past.

Although the above embodiment describes the case of recognizing handwritten kanji characters, the present invention is not limited to this and may be used to recognize characters with many straight line segments, such as handwritten katakana characters.

Furthermore, as a determining method of the determining means 1s, a method has been described in which the degree of similarity is determined from the distance between the stuck midpoint pattern of the input character and the staggered midpoint pattern of the reference character. Alternatively, a method may be used in which the pre-weighted stroke midpoint pattern of the reference character is directly superimposed on the stroke midpoint pattern of the input character to determine the degree of similarity.

As explained above, 1.According to this invention, recognition is performed using the stroke midpoint pattern extracted after line thinning, so it is stable against small fluctuations in the slope of line segments and fluctuations in line width. In addition, this method has the advantage that handwritten characters can be recognized with high accuracy because it is less affected by the combination or separation of character lines, which is a disadvantage of methods characterized by feature points such as one end point or a branch point.

[Brief explanation of the drawing]

Figures 1 and 2 are diagrams for explaining examples of variations in input character patterns, Figure 3 is a block diagram showing an embodiment of the device of the present invention, and Figure 4 is a diagram with thin lines. Figure 5 is a diagram showing an example of an input character pattern, Figure 5 is a diagram showing an example of an input character pattern with segments extracted, Figure 6 is a diagram showing an example of an input character pattern with strokes extracted, Figure 7 is a diagram showing an example of an input character pattern with strokes extracted. FIG. 8 is a diagram showing an example of a midpoint pattern of a standard character. In the figure, 11 is a form, 12 is a scanning means, 13 is a preprocessing means, 14 is a stroke extraction means, and 15 is a determining means. Note that the same reference numerals in the figures indicate the same or similar parts. Agent Makoto Kuzuno (1 other person) Figure 1 Figure 2 Figure 3 Figure 4 Figure 6 Figure 17゜

Claims (1)

[Claims] Character U that recognizes characters recorded on recording media such as forms
The recognition device 1 includes a scanning means for scanning and photoelectrically converting the characters, a preprocessing means for thinning the input character pattern obtained by the scanning means, and a straight line segment from the thinned input character pattern. a stroke extracting means for extracting;
determining means for determining a character using the midpoint pattern of the straight line segment obtained by the stroke extraction means, the stroke extraction means Extract feature points such as end point 9-segment line points and refraction points, find a line segment that connects them, and combine pairs of line segments that connect to feature points other than end points in the same direction to form a straight line. extracting a line segment and recognizing a character using the degree of similarity between the midpoint pattern of the straight line segment and the midpoint pattern of the straight line segment of a reference character pattern stored in advance in the determining means; Character recognition device.