基于双神经网络分类器的脱机手写体汉字识别  

Exploring Recognition of Off-Line Handwritten Chinese Characters Using Double ANN(Artificial Neural Network) Classifier

作　　者：王歌[1] 谢松云[1] 党正[1] 

机构地区：[1]西北工业大学电子信息学院,陕西西安710129

出　　处：《西北工业大学学报》2010年第4期574-578,共5页Journal of Northwestern Polytechnical University

基　　金：国家自然科学基金(304704591);西北工业大学基础研究基金(W018102)资助

摘　　要：脱机手写体汉字识别因其自身的复杂性,系统的实现具有很大的困难。现有的方法,多针对小字符集。为了能在更大的字符集内实现脱机手写体汉字识别,文章结合双重特征提取方法,提出了将双神经网络分类器引入脱机手写体汉字识别。该方法提取汉字字符的2组特征,将2组特征输入双重神经网络进行并行训练,再经过后处理选择最优结果。将该方法与用于小字符集的SVM方法进行了比较,结果表明其识别率明显高于SVM方法,说明双神经网络在脱机手写体汉字识别中有较强的可行性和实用性。To implement the recognition of off-line handwritten Chinese characters is an arduous task.Most of the existing methods have in mind just the small character set.So we combine a double ANN classifier with double feature extraction to implement the recognition of the bigger character set.Our method extracts the double features of one Chinese character and then uses them as inputs to train the double ANNs in parallel.Finally,the post-processing module chooses a more robust result as the recognition output.The comparison between the recognition rates of our method with those of SVM（support vector machine） method,presented in Tables 1 and 2,shows preliminarily that our method is better than the existing SVM method for the recognition of bigger off-line handwritten Chinese character set.

关 键 词：神经网络 分类器 脱机手写体 汉字识别 

分 类 号：TP391.43[自动化与计算机技术—计算机应用技术]