JPH05108882A - Character recognition device - Google Patents

Abstract

(57) [Summary] [Purpose] To improve the recognition rate by reducing the effects of noise and missing graphic features of characters. [Structure] A character area is cut out from a binary image input by an image input unit, and the character area is divided into small areas by the number of vertical and horizontal divisions determined by the aspect ratio of the character area, and black pixels in each small area are divided. The ratio of white pixels and white pixels is obtained as the graphic feature of the character area, and the recognition result is obtained by comparing the graphic feature with the recognition dictionary by a method such as similarity. [Effect] The character area division pattern in the feature extraction is determined by the aspect ratio of the character area. We aim to improve the recognition rate by mitigating the effects of missing figure features and noise.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition device for recognizing characters.

[0002]

2. Description of the Related Art A conventional character recognizing device divides a character area into small areas by defining a circumscribing rectangle of a character as a character area, and fixing a fixed number of divisions in the vertical and horizontal directions. The ratio of pixels was obtained as a graphic feature of the character area.

When the number of divisions in the vertical and horizontal directions is fixed in dividing the character area, "vertical elongated characters" such as "I" and "1" and "-" and "=", etc. In the case of "horizontally long and narrow characters", the amount of information in the vertical and horizontal directions is greatly different, which has a considerable adverse effect on the recognition result.

For example, considering the recognition of'i ', the major feature is that the upper and lower islands are separated,
The feature is information in the vertical direction, and it is difficult to appear unless the number of divisions in the vertical direction is increased and the division width is reduced. Further, since the feature in the horizontal direction is small, unless the number of divisions in the horizontal direction is reduced and the division width is increased, it is easy to be affected by noise or the like.

Therefore, in "i", it is desired to increase the number of divisions in the vertical direction and decrease the number of divisions in the horizontal direction. As long as you do the same division for the characters, it affects either character.

As a result, if a sufficient number of vertical divisions cannot be set for'i ',' i 'is often mistaken for'I' or '1'.

[0007]

As described above, since the conventional character recognition device divides the character region in the feature extraction in the same manner for any object,
This led to a lack of important features and an increase in the influence of noise.

[0008]

SUMMARY OF THE INVENTION The present invention has been made in view of the above problems, and the division pattern of a character area in feature extraction is determined by the aspect ratio of the character area. Characters that are coarse and long and narrow horizontally are divided finely horizontally and vertically to reduce the effects of missing graphic features of characters and noise, and improve the recognition rate.

[0009]

By reducing the effects of noise in the graphic features of characters and noise, it is possible to suppress the mutual recognition of similar characters and the recognition error due to noise, and to improve the recognition rate.

[0010]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described with reference to the accompanying drawings showing one embodiment.

In FIG. 1, reference numeral 1 denotes an image input unit for inputting a document to be recognized as a binary image, and 2 denotes a circumscribed rectangle of a character based on connection information of black pixels in the input binary image. , A character area cutout unit that cuts out a character area one by one, and 3 is determined by the aspect ratio of the character area.
The feature extraction unit 4 divides a character area into small areas by the number of horizontal divisions, and obtains the ratio of black pixels and white pixels in each small area as a graphic feature of the character area. Are compared by a method such as a degree of similarity, and the character identification unit that recognizes the character having the most similar feature as the recognition result of the character area,
Reference numerals 7 to 7 are recognition dictionaries that store the graphic features of all characters to be recognized, and are used properly according to how the character region is divided by the feature extraction unit 3. Also, these recognition dictionaries are prepared in advance. Reference numeral 8 is an internal bus connecting the units 1 to 4, and 9 is an internal bus connecting the units 5 to 7.

With respect to the character recognition apparatus of the present embodiment configured as described above, FIG. 2 shows a flow chart of the entire processing, and FIG. 3 shows an example of character area division. The operation will be described below. The document to be recognized is input as a binary image by the image input unit 1 (process 10). The character area cutout unit 2 obtains a circumscribed rectangle of the character based on the connection information of the black pixels in the binary image, and cuts out each character as a character area (process 11). The feature extraction unit 3 extracts the graphic feature of the character area (processes 12 to 1).
8). The process will be described below.

First, the aspect ratio of the character area is obtained, and the division pattern of the character area at the time of feature extraction is determined according to the value of the aspect ratio (processes 12 to 15). The division is performed by dividing the character area into a board shape, and the number of divisions in the vertical direction and the number of divisions in the horizontal direction are set by the aspect ratio.

In the present description, depending on the aspect ratio, "vertical: 1
2, horizontal: 5 "" vertical: 8, horizontal: 8 "" vertical: 5, horizontal: 12 "
These three divided patterns are separately used to form divided patterns 1, 2, and 3, respectively. Each division pattern is shown in FIG.

The division pattern is as follows according to the aspect ratio. When aspect ratio> 2.0 Divided pattern 1 0.5 ≤ aspect ratio ≤ 2.0 Divided pattern 2 When aspect ratio <0.5 Divided pattern 3 Select a character area with the divided pattern determined by the above. It is divided into small regions (process 16), and the area of the black pixel in each small region is obtained (process 17). In order that the extracted feature does not depend on the size of the character, the obtained area of the black pixel is divided by the area of the small region and normalized (process 18). The normalized value of each small area is used as a graphic feature.

Based on the graphic feature of the character area obtained by the above processing, the character identifying section 4 identifies which character the graphic in the character area corresponds to (process 19 to 2).
3).

First, a recognition dictionary is selected according to the division pattern selected at the time of feature extraction (process 19). Three types of recognition dictionaries are prepared in accordance with the division patterns. The dictionaries A, B, and C are created in advance by extracting features from all the characters to be recognized in the division patterns 1, 2, and 3, respectively. When the division pattern selected at the time of feature extraction is 1, the dictionary A is selected, when it is 2, the dictionary B is selected, and when it is 3, the dictionary C is selected.

Next, the selected dictionary and the graphic feature are compared by a method such as similarity, and the character having the most similar feature is set as the recognition result (processes 10 to 23).

[0019]

As described above, according to the present invention,
By reducing the effect of noise in the graphic feature of a character and the influence of noise, mutual recognition of similar characters and erroneous recognition due to noise can be suppressed, and the recognition rate can be improved.

[Brief description of drawings]

FIG. 1 is a configuration diagram of a character recognition device according to an embodiment of the present invention.

FIG. 2 is an overall flow chart of character recognition processing according to this embodiment.

FIG. 3 is a diagram showing a division pattern of a character area according to the present embodiment.

[Explanation of symbols]

 1 Image Input Section 2 Character Area Extraction Section 3 Feature Extraction Section 4 Character Identification Section 5-7 Recognition Dictionary 8-9 Internal Bus 10-24 Processing in Flow Chart 25-27 Character Area Division Pattern

Claims (1)

[Claims] 1. An image input unit for inputting a document to be recognized as a binary image, and a character circumscribing rectangle is obtained based on connection information of black pixels in the input binary image to determine a character area. The character area cutout part that cuts out each character and the character area is divided into small areas by the number of vertical and horizontal divisions determined by the aspect ratio of the character area, and the ratio of black pixels and white pixels in each small area It is composed of a feature extraction unit which is obtained as a feature, and a character identification unit which compares the extracted features with a dictionary prepared in advance by a method such as a similarity degree and uses a character having the most similar feature as a recognition result of a character region. Character recognizer.