Preclassification for handwritten chinese character recognition by a peripheral shape coding method

摘要

Preclassification is a promising strategy in Chinese character recognition for solving the difficulty of recognizing a very large character set. In this paper, a preclassification technique based on peripheral shape coding is proposed. According to the fact that the peripheral strokes are usually easier to extract and more stable than the strokes in the central portion of a handwritten Chinese character, 37 fundamental stroke patterns called radicals are explored and partitioned into 25 categories. Each category is assigned a code. In order to extract the code from the input characters, some preprocessing stages are first used to obtain the skeleton and feature points of the input character, then an algorithm is proposed to extract the four radicals from its four corners to form a codeword. Based on this codeword, every character can finally be classified into the desired class. Experimental results show that the commonly used 5401 handwritten Chinese characters could be partitioned into 2144 classes, each class containing on an average 2.5 characters. A preclassification accuracy of 94% can be attained for the characters written by a person who had been trained for 1 h, and 98% for a skilled person.