JPS59153268A - Recognizing method of handwriting character - Google Patents

Abstract

PURPOSE:To use an overlapping method without using a structural analyzing method absorbing inclination by detecting the inclination an inputted handwriting character and correcting the shape of the inputted character. CONSTITUTION:A character inclination extracting part 2 reads out a handwriting character pattern from a storage part 1 as a signal 11, outputs the inclination information as a signal 21, extracts and roughly classifies the inclination information, and outputs the candidate kind of character as a signal 22. A pattern correction part 3 reads the inputted handwriting character pattern as the signal 11, corrects the inclination of the inputted character pattern in accordance with inclination information inputted as the signal 21 and outputs the corrected pattern as a signal 31. A character recognition part 4 reads the correction pattern as the signal 31, executes the recognition processing of the candidate character kind sent as the signal 22 and outputs the recognized result as a signal 41.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a handwritten character recognition system, and more particularly to a handwritten kanji recognition system.

In recent years, the development of character recognition technology has been remarkable.As for the recognition of alphanumeric characters, B deviation for handwritten characters and printed characters has also been commercialized, and has been put into practical use.
For those that recognize Japanese characters including hiragana,
As far as single fonts for printed characters are concerned, the development of prototype machines has already been announced.

However, although results are gradually being achieved in the recognition of handwritten kanji, the reality is that it is not very robust against character deformation, which is the inherent problem with handwritten characters.

One of the most problematic types of character deformation is the difference in the inclination of the characters depending on the scribe. Moreover, compared to conventional OCR for alphanumeric and kana characters, the inclination of handwritten kanji is unique to each scribe, and it is extremely difficult to change the writing style that has been learned over many years since childhood. Therefore, a character recognition method that absorbs the inclination of characters for each scribe is required.

An object of the present invention is to provide a recognition method by making the inclination of an input handwritten character steep and correcting the shape of the input character. Although an analytical method has been used, this method allows a superposition method to be used.

The details of the present invention will be explained below using the figures.

FIG. 1 is a block diagram showing an example of the configuration of the present invention.

1 is a quantized character stamp storage unit that stores handwritten character stamps inputted as a signal 10, and 2 is a character inclination extraction unit that reads the handwritten character pattern from the storage unit 1 as a signal 11 and stores the information on the inclination. The candidate character types are output as a signal 21, and the candidate character types are output as a signal 22. 3 is the slam correction part 6D,
The input handwritten character pattern is read in as a signal 11, and the tilt of the input character button is corrected according to the tilt information input as a signal 21, and the corrected button is output as a signal 31. 4 is the character recognition unit p, and the correction pattern is signal 3
1, performs recognition processing based on the candidate character type sent as signal 22, and outputs the recognition result as signal 41.

Among the above-mentioned components, the quantized character stamp storage section l may be a commonly used RAM. Character slant extraction part 2
This can be done by synthesizing existing means such as the ``Character Classification Method'' of Japanese Patent Application No. 57-098823 and the ``Directional Quantization Method of Line Figures'' of Japanese Patent Application No. 57-039011. The baton correction unit 3 scans the input baton in a diagonal direction as shown in FIG. 2(al) and outputs the correction pattern shown in FIG. 2(b). Just use it.

The character recognition unit 4 may use a recognition method using an existing superimposition method, and for example, the method shown in Japanese Patent Application No. 169153/1983 "Character Recognition Method" may be used for candidate character types. .

With the above-described standard, it is possible to realize a handwritten character recognition method that adapts to variations in the inclination of each input character. Note that the character inclination extraction unit 2 performs major classification to reduce the burden on the character recognition unit 4, but if there are only a few character types, the character recognition unit 4 may recognize all character types without performing major classification. Processing may be performed.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an example of the configuration of the present invention, where l is a quantized character slam storage unit, and 2 is a character representative ben vshi Uchihara Susumu 71 Figure 2 m.

Claims (2)

[Claims] (1) In a method for recognizing handwritten Japanese characters such as kanji, a quantized character slam storage unit stores quantized input handwritten character bangs, and the handwritten character bang signal is received from the quantized character bang storage unit. a character slant extraction unit that reads, extracts slant information, and outputs a slant information signal; and a character slant extractor that reads the handwritten character slam signal, simultaneously inputs the slant information signal, and extracts the slant information signal, and outputs a slant information signal. a slam correction unit that outputs a corrected character slam signal that has been corrected; 1. A handwritten character recognition system, comprising a character recognition unit that outputs a signal as a signal, and performs recognition processing adapted to variations in the inclination of the handwritten input character button. (2) The character slant extraction section performs broad classification at the same time as extraction of slant information, outputs slant information and candidate character type signals as a result of the main classification, and the character recognition section inputs a correction pattern signal. The handwritten character recognition method according to claim 1, wherein a signal of candidate character types resulting from the major classification is inputted, and a recognition process is performed on the candidate character types.