摘要
采用基于识别的分割方法进行手写数字串分割.在识别的过程中,运用反例样本估计分类器参数,实验数据表明,这种运用反例样本训练的分类器与没有经过反例样本训练的分类器相比,将提高拒识率到19%左右,从而保证了较高的识别率,验证了只有经过反例训练的分类器的输出结果才是可信赖的.
This paper used the recognition-based Method to solve the segmentation problem of handwritten numerical strings . In the segmentation process, to get classifier with better performance, negative data must be the necessary trained samples. The experiment results show that this method with negative data can get better refuse rate, which reaches 19 percent. During to the increase of refuse rate, the recognition accuracy also increases,validate that the outputs of the classifier trained with negative data is reliable.
出处
《武汉大学学报（理学版）》
CAS
CSCD
北大核心
2007年第3期301-304,共4页
Journal of Wuhan University:Natural Science Edition
基金
湖北省科技攻关计划资助项目(2003BDSP004)
关键词
数字串分割
反例样本
分类器
numerical strings segmentation
negative data
classifier
