JPS62145485A - Character recognition system - Google Patents

Abstract

PURPOSE:To correctly segment a character without reducing the recognition speed and to enhance a recognition rate by providing the plural systems of character segmenting parts and recognition parts. CONSTITUTION:The character segmenting parts 21, 31 respectively segment simultaneously according to different segmenting methods with respect to a character sequence pattern stored in a line buffer 1. The recognition part 22 recognizes the character pattern segmented in the segmenting part 21 and informs a candidate character and its similarity to a result decision part 4. Similarly, the recognition part 32 performs the processing and informs to the decision part 4. The decision part 4, when both systems complete the recognition of the character sequence pattern of one line, decides whether either one of the recognition results from both systems is reasonable or not based on the similarity of the respective candidate characters and outputs the final recognition result. Thereby, the correct character segmenting is performed without reducing the recognition speed and the recognition rate can be enhanced.

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field] The present invention relates to a character recognition method for reading and recognizing character images on a form.

(Prior art) In character recognition in a character recognition device, the L!!
In order to improve the recognition rate, accurate character extraction is essential.Normally, character extraction is done simply by using a fixed pitch or based on vertical projection, but it can be done on any form or character. The current situation is that there is no one that can be said to be optimal for the image.

Therefore, it is necessary to check whether the cut out character pattern has been accurately recognized, that is, the validity of the character, taking into consideration the recognition result.

However, conventional character extraction is performed independently as preprocessing for character recognition, and it is rare to organically link the two to perform character recognition.

If an attempt was made to perform accurate extraction while recognizing, there would be a problem that the recognition speed would decrease.

(Purpose) The purpose of the present invention is to provide a character recognition method that improves the accuracy of extraction and has a high recognition rate even for character strings that include characters that are difficult to extract, such as single-separated characters and contact characters. It is in.

〔composition〕

The present invention provides a plurality of systems of character extraction units and recognition units, and recognizes the same character string pattern by simultaneously performing character extraction using different extraction methods in the plurality of systems. By determining a valid recognition result for each character from the recognition results of each system at the stage when recognition is completed, a character recognition device with a high recognition rate can be realized without reducing recognition speed.

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram of an embodiment of the character recognition method of the present invention, which includes the line buffer 1, the first character extraction/recognition system 2 of the character extraction section (1) 21 and the recognition section (1) 22, and one character. A second character extraction/recognition system 3 of the extraction section (2) 31 and the recognition section (2) 32.

It consists of a result determination section 4.

The character string pattern for one line on the form is converted into a binary pattern and stored in the line buffer 1. For the character string pattern of this line buffer 1, the character cutting section (1) 21
and character segmentation unit (2) 31 simultaneously execute character segmentation using different segmentation methods. For example, the one-character cutting section (1) 21 performs constant pitch cutting using a prespecified pitch, and the cutting section (2) 31 performs cutting based on vertical projection. Since these character extraction methods are well known, their details will be omitted. Recognition unit (1) 22
After extracting the features of the character pattern cut out by the 9-character extraction unit (1) 21, it is compared with a feature dictionary prepared in advance, and the candidate characters and their similarity are notified to the result determination unit 4. do. Similarly, the recognition unit (2) 32 also performs recognition processing on the character pattern cut out by the character cutout unit (2) 31, and notifies the result determination unit 4 of candidate characters and their degrees of similarity. In the result judgment section 4, 1-car system 2.3 is 1
When recognition of the character string pattern for a line is completed, based on the degree of similarity of each candidate character, it is determined which system the recognition result from is valid, and the final recognition result is output.

FIG. 2 shows a block diagram of another embodiment of the present invention, in which each time recognition for one character is completed, the result determination control unit 4 determines which system of recognition results is more valid. , and is fed back to each character segmentation section 21, 31 to be reflected in the next character segmentation, thereby increasing the accuracy of character segmentation.

For example, one character cutting part (1) 21 and character cutting part (2)
31 both cut out characters using vertical projection, but the process of concatenating separated pans is different for the single character patterns "ii" and "ha", as shown in Figure 3. It is assumed that the character cutting portion (1) 21 is cut out as shown in (1) in the same figure, and the character cutting portion (2) 31 is cut out as shown in (2) in the same figure. In this case, the result determination control unit 4 determines that FIG. 3 (1) is more appropriate for the third character based on the similarity from the recognition unit 22. to correct the cutting position from b to a. As a result, even though the cutting unit (2) 31 incorrectly cuts out the third character, it can carry out the secondary cutting process of the fourth character without being affected by the incorrect cutting.

In addition, for the pattern "I", the cutout part (1) 21
However, the cutting unit (2) 31 performs the correct cutting. In this case, the result judgment control unit (4) sends a signal to correct the cutting position from C to d for the cutting unit (1) 21. It is indicated via line 41. In addition, each character cutting section 2
The character extraction method of 1.31 may be communicated to the result determination control section 4 in advance through the signal line 23.33.

In the embodiments shown in FIGS. 1 and 2, there are two systems for single character extraction/recognition, but it goes without saying that three or more systems may be used.

〔effect〕 As is clear from the above description, one aspect of the present invention is as follows.

1 by providing multiple systems of character extraction units and recognition units.
Even for character strings that include characters that are difficult to extract, such as separated characters and touching characters, the recognition speed is not reduced.
Accurate character extraction can be performed to increase recognition rate.

[Brief explanation of drawings]

FIG. 1 is a block diagram of one embodiment of the present invention, FIG. 2 is a block diagram of another embodiment of the present invention, and FIG. 3 is a diagram for explaining the operation of FIG. 2. 11. , line buffer, 2...first character extraction/
Recognition system, 3... Second character extraction/recognition system, 4...
-Result judgment (control) section, 21.31...Character cutting section, 22.32...Recognition section. Representative Patent Attorney Makoto Suzuki -・, Near 1 ・I-+U・ Figure 1 Figure 2

Claims (2)

[Claims] (1) In a method for cutting out and recognizing character patterns character by character from a character string pattern, multiple systems of character cutting sections and recognition sections are provided, and character patterns are cut out using different cutting methods for each system for the character string pattern. A character recognition method is characterized in that a character recognition method is used to determine which system's recognition result is valid. (2) A character recognition method according to claim 1, characterized in that character segmentation information based on a system determined to be appropriate is transmitted to a character segmentation unit of another system.