JPS5822779B2 - Character pattern classification method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a classification method for character patterns, particularly for printed kanji patterns obtained by photoelectric conversion.

This invention relates to a character pattern classification method in which character patterns are standardized by their circumscribed areas and classified based on several characteristics of the patterns, in order to cope with changes in character shapes due to differences in fonts.

Traditionally, the classification method for character patterns including kanji is
For example, classification based on information surrounding a character pattern cut out in a circumscribed rectangular area, classification based on information on the complexity of a character pattern, etc., is known for targeting character shapes of a single font.

Of these, the former classification method extracts an area of a certain width inside the character frame, and uses information about the character lines contained therein and changes in that information when the characters are made thicker.

However, classification based on information on character lines included in an area of a certain width has the disadvantage that it is easily affected by changes in character shape that occur in printed kanji patterns with different fonts, such as expansion and contraction of character lines. Although it is possible to stably extract white area information by classifying based on changes in information in character frame portions, there are problems in that the processing becomes complicated and the amount of memory required increases.

On the other hand, in the latter method, that is, classification using a complexity index defined from the sum of the amount of vertical and horizontal character spread in a character pattern and the length of character lines, the processing is uneven and complicated. This has the drawback that it is easily affected by pattern deformation in parts that are essentially unrelated to kanji recognition, such as ``serifs'' and ``scales'' that exist in printed kanji patterns.

In order to solve these problems, the present invention classifies character patterns using rough complexity information and rough outline shape information of printed kanji patterns, which is a simple process. The purpose is to provide a classification method that requires less memory.

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 classification method according to the present invention.

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

This photoelectrically converted signal is compared with a reference value in the binarization circuit 3, and the white area is IT Ojj and the black area is 2°1''.
The value is converted to a quantized character pattern.

4 is a character pattern circumscribing frame detection circuit, and an example of its operation is shown in FIGS. 2A and 2B.

That is, as shown in FIG. 2A, the tentacle 12 is extended toward the center of the frame 11 of the binarized character pattern, and when it first encounters a character part, the frame 13 of that side is performed.

Similar processing is performed on the other sides to detect the circumscribing frame 14 of the character pattern as shown in FIG. 2B.

Information on the detected character pattern within the circumscribed frame is stored in the storage device (
memory) 5.

The character patterns stored in the storage device 5 are classified into the following two formats.

One method is to extract features using a coarse mesh pattern creation device 6 and classify the features using a classification device 7. An example of this operation is shown in FIG.

FIG. 3A shows a character pattern cut out using the circumscribing frame 14.
As shown in FIG.
-1, 15-2. .......

15-n, the area of the character part existing in each mesh area is counted, and this is normalized by the area of the entire character to create a feature for classification.

Figure C is an example of a feature table extracted in this manner, in which the normalized value is multiplied by a constant value and is expressed as an integer.

Based on the feature table created in this manner, the classification device 7 performs matching with the previously stored feature tables for each character to classify character patterns.

The other method is to use a coarse peripheral pattern creation device 8 to determine the characteristics of the outer contours of characters, and based on these characteristics, a classification device 9 is used to classify them.
An example of its operation is shown in FIG.

FIG. 4A shows a character pattern cut out using the circumscribing frame 14.
As shown in B and C of the same figure, this character pattern area is
First, the circumscribed frame is roughly divided by the coarse peripheral pattern creation device 8.

Figure 4B is a rough division of the horizontal circumscribing frame, Figure 4C
is roughly divided into vertical circumscribed frames.

Next, each portion 16-1, 16-2 of the divided circumscribed frame.
......, count the area of the white area (shaded area in the figure) in the direction of the character frame on the opposite side from 16-n until it meets the character part, and calculate this as the area of the entire character pattern area. to create features for classification.

The figure is an example of a feature table obtained for the four circumscribed frames in this way.

Based on this feature table, the classification device 9 performs matching with the previously stored feature tables for each character to classify the character pattern.

FIG. 5 shows an example of the classification results of character patterns by the classification devices 7 and 9.

In other words, for the unknown input character ``mizu'', as a result of classification using a coarse mesh pattern, in order of consistency in matching, ``water'', ``ice'', ``ki'', ``ei'',
"Books" are classified.

Similarly, by classification based on coarse peripheral patterns, "water"
, "tree", "ice", "book".

"English" is selected.

The classification device 10 selects the union of a plurality of categories classified using these two methods, and performs general classification of character patterns.

In the embodiment, the classification devices 7, 9, and 10 are shown separately, but it is also possible to replace some or all of them with one classification device.

As explained above, in the character pattern classification method according to the present invention, character patterns are classified using information on the rough complexity of the entire character pattern and rough information on the outline shape of the character pattern 7. This method has the advantage that it is possible to efficiently classify patterns such as character shape deformations, especially patterns in which the shape of the character pattern differs due to different fonts, such as multi-font printed Chinese characters.

Further, since the features created in the two classification formats can have the same number, properties, etc., a classification device can also be used in common.

Furthermore, depending on the input character pattern, it is possible to identify the character pattern by finding a common category with a high degree of consistency selected by the classification device.

[Brief explanation of drawings]

FIG. 1 is a diagram showing an embodiment of the character pattern classification method according to the present invention, FIG. 2 is a diagram explaining the operation of the circumscribing frame detection circuit of FIG. 1, and FIG. 3 is a diagram showing the coarse mesh pattern of FIG. 1. Figure 4 is a diagram for explaining classification based on coarse peripheral patterns in Figure 1;
FIG. 5 is a diagram showing an example of the classification results of character patterns according to the present invention. 1... Input character pattern, 2... Scanning/photoelectric converter, 3... Binarization circuit, 4...
...Circumscribing frame detection circuit, 5...Storage device, 6...
...Coarse mesh pattern creation device, 7...
- Classification device, 8... Coarse peripheral pattern creation device, 9... Classification device, 10... Classification device.

Claims (1)

[Claims] A first means for framing a binary character pattern using a circumscribed rectangular area of the character, and dividing the character area obtained by the first means into rectangular areas with a coarse mesh. a second means for counting the area of the character portion existing in each region and classifying the character pattern based on this information; and a second means for roughly dividing the circumscribing frame of the character region obtained by the first means. , this minute. a third means for counting the area of a white region from each divided portion in the direction of the character frame on the opposite side until it meets a character portion, and classifying character patterns based on this information; and a fourth means for classifying an unknown input character pattern by combining the classification results obtained by the means.