北京邮电大学学报

  • EI核心期刊

北京邮电大学学报 ›› 2006, Vol. 29 ›› Issue (s2): 118-121.doi: 10.13190/jbupt.2006s2.118.cheng

• 论文 • 上一篇    下一篇

基于余弦整形变换的手写汉字训练样本生成方法

陈 光, 张洪刚, 郭 军   

  1. 北京邮电大学 信息工程学院, 北京 100876
  • 收稿日期:2006-08-31 修回日期:1900-01-01 出版日期:2006-11-30 发布日期:2006-11-30
  • 通讯作者: 陈 光

Learning Pattern Generation for
Handwritten Chinese Character Using
Pattern Transform Method with Cosine Function

CHEN Guang, ZHANG Hong-gang, GUO Jun   

  1. 1. School of Information Engineering, Beijing University of Posts and Telecommunications, Beijing 100876, China
  • Received:2006-08-31 Revised:1900-01-01 Online:2006-11-30 Published:2006-11-30
  • Contact: CHEN Guang

摘要:

提出了一种新的基于余弦整形变换的手写汉字训练样本生成方法,该方法可以有效提高现有手写汉字识别系统的识别性能。通过对原始训练样本进行基于余弦整形变换的样本生成和样本筛选,在一定程度上缓解了某些条件下训练样本不足的问题,增强了识别器的扩展能力和鲁棒性。在HCL2000数据库上进行的实验表明,该方法可有效提高汉字分类器的识别率。

关键词: 模式识别, 手写汉字识别, 余弦整形变换, 训练样本生成

Abstract:

Based on pattern transform method with cosine function, a new pattern generation method for handwritten Chinese characters was proposed, which improved the performance of recognition system for handwritten Chinese characters. By using transform method with cosine function, the number of learning patterns was increased. Patterns generated artificially in this way were selected using pattern selection method and the patterns unsuitable for learning were discarded. The recognition experiment on HCL2000, a handwritten Chinese character database, showed that our method improved the recognition precision of whole system.

Key words: pattern recognition, pattern generation, Chinese character recognition, pattern transform with cosine function

中图分类号: