JPH0253831B2 - - Google Patents

Description

[Detailed Description of the Invention] (A) Technical Field of the Invention The present invention provides a character recognition device, particularly a character recognition device that separates kanji into, for example, “hen” and “tsukuri” to improve recognition accuracy. Character recognition that extracts character internal line segments that represent white areas interposed by character strokes, determines their series, and separates one character into partial patterns based on the manner in which the series occurs. It is related to the device.

(B) Background and problems of the technology In the process of recognizing kanji characters, the characters to be recognized are first roughly classified by some known means, and then the characters to be recognized are classified into characters such as "hen" and "crown." It is being considered that the recognition accuracy can be improved as a whole by separating into partial patterns such as ``'' and ``zukuri'' and recognizing each of them.

However, in the case of kanji, especially handwritten kanji, for example, the words ``hen'' and ``tsukuri'' may partially touch each other, making it difficult to separate them properly. This has the drawback of lacking versatility, as it is necessary to use different processing methods depending on the type of partial pattern.

(C) Object and Structure of the Invention The present invention aims to solve the above-mentioned problems, and the character recognition device of the present invention scans the kanji characters to be recognized, extracts the features, and extracts the features corresponding to the standard kanji characters. A character recognition device that improves recognition accuracy by separating a single character into partial patterns when determining the category of the kanji character to be recognized by comparing the characteristics with the contents of a dictionary in which the characteristics are stored. , a series of horizontal character internal line segments representing white areas sandwiched by character strokes when searching for the recognition target Kanji character in the horizontal direction, and character strokes when searching for the recognition target Kanji character in the vertical direction. At the same time, extract a series of vertical character internal line segments that represent the white area sandwiched between characters, and extract a series of character internal line segments when separation occurs for each category character that can be separated within one character. The character is stored in a dictionary for partial pattern separation, and a single character is separated into partial patterns by comparing the extracted series of character internal line segments with the contents of the dictionary. This will be explained below with reference to the drawings.

(D) Embodiments of the Invention FIGS. 1A and 1B are explanatory diagrams each explaining the concept of character internal line segments and partial pattern separation according to the present invention, and FIG. 2 is a diagram for partial pattern separation used in the present invention. An explanatory diagram illustrating the contents of an example of a dictionary,
FIG. 3 shows the configuration of an embodiment of the present invention.

When a handwritten kanji character 1 as shown in FIG. 1A is given, a scan 2 is performed in the direction of the arrow shown in the figure, and (i) point A, which first reaches the black area from the left white area of the background, is extracted.

(ii) Next, extract the point a that reaches the white area from the black area.

(iii) Next, extract the point B that reaches the black area from the white area.

(iv) Next, extract the point b that reaches the white area from the black area.

(v) Finally, extract the point n that reaches the right white area of the background from the black area.

Do it like this. Then, the center point between point a and point B extracted in this way (temporarily set as P1), point b and point N
Find the center point (temporarily assumed to be P2) between the two points.

Under this state, in the case shown in FIG. 1, points P1, P2, . . . are obtained corresponding to the vertically arranged scanning lines.
. . are connected in the vertical direction, and horizontal character internal line segments L1, L2, . . . as shown in FIG.
Extract... At this time, if the horizontal positions of the points corresponding to, for example, the point P1 obtained in correspondence with the two vertically arranged scans are separated by a threshold value or more, the line segment L1 etc. is considered to be discontinuous. When the starting end and ending end of each line segment are found, the illustrated line segments L1, L2, . . . are obtained by connecting both ends with straight lines. Further, cases in which the start and end ends are sealed by a black area of the character (black circles in the figure) and cases in which they are not sealed (white circles in the figure) are extracted separately.

Extraction of character internal line segments as described above can be similarly obtained when a character is scanned in the vertical direction.

The horizontal character internal line segment obtained when scanning in the horizontal direction is the line segment L shown in Figure 1 A and B.
1, L2, . . . as shown in FIG. 1A, there may be a series of line segments L1, L2, L3,
Furthermore, it is clarified that if there is a series such as line segments L1, L2, and L3 as shown in FIG. 1B, the character "kyo" can be separated into left and right partial patterns.

In this way, one character, for example, the handwritten character ``kyo''
Various series of horizontal character internal line segments that can be separated into partial patterns are investigated, and each of these series is registered in a partial pattern separation dictionary as shown in FIG.

FIG. 2 shows various "sequences of character internal line segments" that appear when the character "kyo" is separated into left and right parts. Generally, a plurality of types of series are registered, such as the illustrated series corresponds to the case shown in FIG. 1A, the illustrated series corresponds to the case illustrated in FIG. 1B, and the illustrated series...

FIG. 3 shows the configuration of an embodiment of the present invention.
Reference numeral 3 in the figure is an input character, for example, A in Figure 1.
Characters 4 as shown in FIG.
5 is a partial pattern separation determination circuit which determines whether the characters belonging to the candidate character group can be separated into left, right, upper and lower partial patterns by known means. What determines
6 is a device that extracts horizontal character internal line segments and obtains their series; 7 is a device that extracts vertical character internal line segments and obtains their series; 8 is a dictionary for partial pattern separation; and 9 is a line segment search circuit. 10 is a partial pattern dictionary in which recognition features for each partial pattern are stored. 11 is a determination circuit which determines what shape the determined partial pattern has based on the contents of the partial pattern dictionary 10, recognizes the input character, and outputs a character code. It represents what you do.

The input characters are sorted by the major classification circuit 4, for example,
It is determined that the character belongs to a candidate character group such as numeral, san, ...). And the input character is left, right or top
The partial pattern separation determination circuit 5 determines whether the partial pattern can be classified into the lower categories. If separable, the above-mentioned series of character internal line segments are extracted by circuits 6 and 7 as described above.

This extraction result is compared with the contents of the partial pattern separation dictionary 8 to obtain a partial pattern. The partial pattern is led to a determination circuit 11, which determines what shape the obtained partial pattern has based on the contents of the partial pattern dictionary 10. Then, the determination circuit 11 knows that, for example, the characters shown in FIG. ”.

Note that conventionally known features can be used as the feature amounts used in the major classification circuit 4, partial pattern separation judgment circuit 5, judgment circuit 11, etc. shown in FIG. Point A
It is possible to use the contour left line segment series obtained when the points n are connected in the vertical direction, and the contour right line segment series obtained when the points n are connected in the vertical direction. When these features are utilized, they can be extracted together by a process similar to the process for extracting the character internal line segment series described above.

(E) Effects of the invention As explained above, according to the present invention, it is easy to separate kanji into multiple partial patterns,
It becomes possible to improve character recognition accuracy.

[Brief explanation of drawings]

FIGS. 1A and 1B are explanatory diagrams respectively explaining the concept of internal character line segments and partial pattern separation according to the present invention, and FIG. 2 shows the contents of an embodiment of a dictionary for partial pattern separation used in the present invention. An explanatory diagram to explain,
FIG. 3 shows the configuration of an embodiment of the present invention. In the figure, 1 is the kanji to be recognized, 2 is the scanning line, 4 is the major classification circuit, 5 is the partial pattern separation judgment circuit, and 6
Reference numeral 1 represents a horizontal line segment extraction circuit, 7 a vertical line segment extraction circuit, 8 a partial pattern separation dictionary, 9 a line segment search circuit, 10 a partial pattern dictionary, and 11 a determination circuit.

Claims (1)

[Claims] 1 Scan the kanji characters to be recognized and extract the features,
When determining the category of the kanji character to be recognized by comparing it with the contents of a dictionary that stores features corresponding to standard kanji characters, the recognition accuracy is improved by separating a single character into partial patterns. In a character recognition device configured as shown in FIG. In addition to extracting a series of vertical internal character line segments that represent the white area sandwiched between character strokes during the search, the above character is extracted for each of the category characters that can be separated within one character. A series of internal line segments is stored in a dictionary for partial pattern separation, and a single character is separated into partial patterns by comparing the extracted character internal line segment series with the contents of the dictionary. A character recognition device characterized by: