摘要
从无限制的手写体数字的结构出发,提出“有效行”特征的概念及其提取算法。该特征具有维数小、平移不变、字符小角度旋转不变等特点。建立了相应的字符特征库,利用BP网络对样本字符进行研究,通过大量的手写体数字识别测试表明:该方法在识别速度上优于利用单一的矩特征、小波特征等传统识别方法;在误识率方面也优于一般的单一识别方法。
The concept and extracting method of the effective-rows feature, which possesses low dimension, shift-invariance and rotation-invariance under little angle is proposed on the basis of the structure of unconstraint and handwritten numerals in this paper. Its features is also constructed. According to the result of widely verifying recognition BP network, it is demonstrated that its recognitionspeed is superior to the traditional recognition methods only based on the feature of moment or wavelet and its recognition-extract is higher than some conventional methods.
出处
《电子科技大学学报》
EI
CAS
CSCD
北大核心
2001年第3期287-291,共5页
Journal of University of Electronic Science and Technology of China