JPS5822781B2 - Character pattern recognition processing method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention uses a character pattern recognition processing method, particularly for printed kanji patterns obtained by photoelectric conversion, in order to deal with changes in character shape due to differences in fonts. The present invention relates to a character pattern recognition processing method in which two complementary features of a character pattern are extracted and classified, and the classification process is hierarchized to perform identification.

Conventional hierarchical recognition methods for character patterns, including kanji, target printed kanji in a single font, apply blurring to the photoelectrically converted unknown input character pattern, and divide it into coarse mesh rectangular regions. There is a known method for recognizing character patterns by classifying them based on their degree of similarity, and by dividing the rectangular regions into smaller areas to create a hierarchy.

However, in this recognition method, the characteristics of the character pattern used are unique, so not only does the recognition accuracy decrease even with slight changes in character shape caused by differences in fonts, but it is difficult to achieve high accuracy. This method has the disadvantage that a complicated process for accurate positioning is required.

Another problem is that the blurring process complicates the processing process.

Furthermore, there are known methods for classifying kanji patterns based on their characteristics, such as using status codes around kanji patterns, complexity indexes, etc., but the features of kanji patterns extracted by these methods are based on unknown input patterns. Since the kanji patterns are uniquely extracted from the kanji characters, it is possible to roughly classify the kanji patterns, but it is impossible to achieve the level of identification through hierarchization.

In order to solve these problems, the present invention has two complementary features of a character pattern: a feature of the general complexity of the character focusing on the character part, and a feature of the rough shape of the character focusing on the part other than the character part. This method realizes efficient recognition by using the features and by layering the patterns representing the features. It is intended to provide.

Hereinafter, the contents of the present invention will be explained in detail with reference to the drawings.

FIG. 1 shows a block diagram of an embodiment of a character pattern recognition processing method according to the present invention.

In the figure, 1 is an input character pattern, which is scanned by a photoelectric converter 2 to obtain a signal proportional to the shading of the character pattern.

This photoelectrically converted signal is compared with a reference value in a binarization circuit 3, and converted into a binary quantized character pattern in which non-text white areas are 0'' and text areas, that is, black areas are 1''.

Reference numeral 4 denotes a character pattern circumscribing frame detection circuit, which extends its tentacles toward the center from each side of the binarized character pattern, and detects the circumscribing frame of that side when it first encounters a character part.

The information of the part surrounded by each circumscribed frame detected in this way is again set as a character pattern, and the memory circuit 5
Remember it.

Reference numeral 6 denotes a hierarchical recognition device which constitutes the main part of the present invention, and a block diagram of one embodiment thereof is shown in FIG.

This figure is an example of a recognition device layered into three layers.

. The hierarchical recognition device 6 recognizes two complementary features of the character pattern, namely, the complexity and positional characteristics of the character pattern focusing on the black area, and the shape and positional characteristics of the character pattern focusing on the white area. The recognition result 7 is obtained by efficiently classifying and identifying character patterns by classifying them into hierarchies based on the level of detail of the information in character patterns that extract the features to be used.

The character pattern stored in the memory circuit 5 is first converted into a characteristic pattern by a black pattern creation circuit 8 and a white pattern creation circuit 9 each having a roughness of approximately 4×4.

FIG. 3 shows an example of the operation for creating a 4×4 coarse black pattern.

Figure 3A shows the character pattern cut out using the circumscribed frame and the character pattern that has been memorized, as shown in Figure 3B.
Divide into 4 coarse rectangular mesh areas, count the area of the character part, that is, the black area, that exists in each mesh area, and normalize this with the area of the entire literature department to create a feature pattern for classification. .

C in the figure is an example of this characteristic pattern, which is a normalized value multiplied by -4 constant and expressed as an integer value.

In this way, the coarse black pattern assumes that the character pattern is characterized solely by the presence of strokes, and uses the information about the character part contained in each area divided into coarse meshes to determine the differences in the rough areas of the character pattern. It represents the characteristics of complexity due to

FIG. 4 shows an example of the operation of the 4×4 coarse white pattern creation circuit 9.

Figure 4A shows a character pattern cut out using a circumscribing frame.The circumscribing frame of this character pattern is roughly divided as shown in Figures B and C, and each part of the divided circumscribing frame is directed toward the opposite side of the circumscribing frame. The area of the non-character part, that is, the white region, is counted until the character part is encountered, and this is normalized by the area of the white region in the entire character pattern region, to create a characteristic pattern for classification.

The figure is an example of a characteristic pattern obtained in this way.

In this way, the coarse white pattern focuses on information on non-character areas seen from the periphery of the character pattern, and expresses the characteristics of the outline shape due to the rough differences in the areas of the character pattern.

Next, with respect to the feature patterns of the unknown input character patterns created by these two feature pattern creation circuits 8 and 9, the matching circuit 10 uses the previously stored character patterns for each category of characters. Matching is performed with the two feature pattern tables 11 and 12, and a certain number of categories are selected from those with the largest values as candidate categories for the unknown input character.

Next, roughly 8×8 black and white patterns are created for the candidate categories extracted in this way.

Reference numeral 13 denotes an 8×8 coarse black pattern forming circuit, an example of its operation is shown in FIG.

Figure A shows a character pattern cut out using a circumscribing frame, which is divided into 8x8 rough rectangular mesh areas as shown in Figure B, and information on the character parts present in each mesh area is used as shown in Figure B. Create a characteristic pattern as shown in C.

On the other hand, 14 is a roughly 8x8 white pattern creation circuit.
An example of its operation is shown in FIG.

For the character pattern shown in A of the same figure, B and C of the same figure
As shown in the figure, the four circumscribing frames are divided into eight parts, and the area of the white area from each part of the divided circumscribing frames to the first encounter with the character part is counted. When calculating the area from
(This will be referred to as the next white pattern; on the other hand, the pattern created by the encounter with the first character portion will be referred to as the primary white pattern), and an 8×8 feature pattern is created in total.

The figure is an example of creating a secondary white pattern from the upper frame.

The matching circuit 1
5, the known characters are compared with two feature pattern tables, and candidate categories are further narrowed down.

Furthermore, for the candidate categories extracted in this way, a black pattern and a white pattern divided into about 16×16 are created.

In the black pattern creation circuit 18, the character pattern area is
By dividing into 16, the white pattern creation circuit 1
In step 9, the circumscribed frame of the character pattern is divided into 16 parts, and a fourth-order white pattern is obtained from 1, similarly to the 8×8 white pattern creation method, thereby creating characteristic patterns each having a desired amount of information.

Using the two feature patterns created in this way, the matching circuit 20 performs matching against the two feature pattern tables 21 and 22, and finally finds one candidate category to identify the unknown input character pattern. The recognition result 7 is obtained.

As explained above, the character batashin recognition processing method according to the present invention focuses on two complementary features of a character pattern, that is, the complexity of the character structure focusing on the character part, and the feature of the complexity of the character structure that focuses on the character part, and the character part that focuses on the non-character part. Characteristics of character shapes,
In addition, character patterns are recognized in a hierarchical manner depending on the features used, that is, the degree of detail of character pattern observation. For patterns that cause differences in pattern shape,
It has the advantage of being able to perform highly accurate and efficient recognition.

In addition, the creation of feature patterns is easy because it can be achieved only by counting processing, and the feature patterns created by two different methods have the same size, properties, etc., so it is easy to create feature patterns at each stage of the hierarchy. Another advantage is that the matching process can be realized by a common circuit.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing an embodiment of the character pattern recognition processing method according to the present invention, FIG. 2 is a diagram showing an embodiment of the hierarchical recognition device which is the main part of FIG. 1, and FIGS. 3 and 5. 4 and 6 are explanatory diagrams for explaining the operation of the black pattern generation circuit, and FIGS. 4 and 6 are explanatory diagrams for explaining the operation of the white pattern generation circuit. In the figure, 1 is an input character pattern, 2 is a scanning/photoelectric converter, 3 is a binarization circuit, 4 is a circumscribing frame detection circuit, 5 is a storage circuit, 6 is a hierarchical recognition device, 7 is a recognition result, and 8
, 13.18 is a black pattern creation circuit, 9, 14.22 is a white pattern creation circuit, 10.15.20 is a matching circuit, 1
Reference numerals 1, 16, and 21 represent feature pattern table storage circuits for black patterns, and reference numerals 12, 17, and 22 represent feature pattern table storage circuits for white patterns.

Claims (1)

[Claims] 1. A means for scanning and photoelectrically converting a character pattern. In a character pattern recognition processing method having means for binarizing information of a photoelectrically converted character pattern, and a storage means for storing the binarized character pattern as a frame in a circumscribed rectangular area of the character. , a literature feature pattern creation means that focuses on the two-character part of the character pattern obtained by the storage means and creates a pattern representing the structure and shape characteristics of the character part; and a character pattern obtained by the storage means. 5 outline shape feature pattern creation means for creating a pattern representing the outline shape of a character based on the features of areas other than character parts; and known character patterns prepared separately from the patterns created by the above two pattern creation means. The pattern creation and matching steps are stratified according to the degree of fineness of the pattern observation, and the processing is executed hierarchically in order to detect unknown input character patterns. 1. A character pattern recognition processing method, comprising: means for recognizing a character pattern.