摘要
提出了一种基于单字单网的手写体汉字识别纯神经网络的多分类器集成方案,并通过实验证明用该方案实现的神经网络集成系统性能均比任一个神经网络单分类器都好,对1 000种不同的手写体汉字的1 000×10个字进行测试,集成后的识别率最高达到95.22%,比单分类器的识别率高出5.0%-8.7%。
This paper proposes a scheme for handwritten Chinese character recognition (HCCR), which is made up of pre-classifiers and model integrated, and both use neural networks. Experiments show that the proposed scheme can classify large number of catalogs well. For example, testing 1 000×10 characters, the maximal recognition rate is 95.22%, 5.0%-8.7% higher than that of advisual classifer.
出处
《计算机工程》
CAS
CSCD
北大核心
2004年第16期151-152,共2页
Computer Engineering
基金
国家自然科学基金资助项目(60275005)
广东省自然科学基金资助项目(011611)
关键词
手写体汉字识别
多分类器集成
神经网络
Handwritten Chinese character recognition
Multi-classifers integrated
Neural network
