Techniques for Solving the Large-Scale Classification Problem in Chinese Handwriting Recognition

Abstract

Given the large number of categories, or class types, in the Chinese language, the challenge offered by character recognition involves dealing with such a large-scale problem in both training and testing phases. This paper addresses three techniques, the combination of which has been found to be effective in solving the problem. The techniques are: 1) a prototype learning/matching method that determines the number and location of prototypes in the learning phase, and chooses the candidates for each character in the testing phase; 2) support vector machines (SVM) that post-process the top-ranked candidates obtained during the prototype learning or matching process; and 3) fast feature-vector matching techniques to accelerate prototype matching via decision trees and sub-vector matching. The techniques are applied to Chinese handwritten characters, expressed as feature vectors derived by extraction operations, such as nonlinear normalization, directional feature extraction, and feature blurring.