摘要
本文介绍一种基于左右轮廓特征识别打印体邮政编码数字的新方法。字符的左右轮廓特征经归一化处理后形成两组独立的特征值，再将它们用线段矢量表示以形成关于轮廓特征的线段描述，并依此对字符进行分类，然后将两个分类器的结果组合起来实现对字符的识别。我们对从真实信函上采集到的８３７４２个打印体邮政编码数字进行测试，实验结果表明具有识别率高（９７．９３％）、错误率低（０．０９１％）等优点。此外本方法在提取特征的同时进行去噪声、归一化等处理，并采用判决树方法，因而识别速度很快，满足了实时识别的要求。本方法已在中比合作开发的ＯＶＣＳ自动信函分拣机中得到应用，实践证明达到设计要求。
This paper presents the recognition of printed zip code numerals based on the numeral left and right profiles.Two group features generated by utilizing the left and right profiles are normalized and described with line vectors,then two classifiers are employed on the two group line vectors.The results of the two classifiers are integrated to recognize the numeral.It is tested with 83742 printed zip code numerals taken from live mails,the results show that the recognition method has good performance for high recognition rates(97.93%) and very low error rates( 0.091% ).Noise removimg and normalizing are processed while extracting profiles,and the decisiontree chassifiers are used,so the recognition speed is very fast and suffices to the requirement of real time recognition.It is sucessfully applied in the OVCS mail sorting machine developed by sino Belgium cooperation,which has proved that the design request and use value are achieved.
出处
《通信学报》
EI
CSCD
北大核心
1997年第7期93-96,共4页
Journal on Communications
基金
邮电部科研课题
中比合作开发项目
关键词
打印体
邮政编码数字
左右轮廓
数字识别
printed zip code numerals,left and right profiles,features,numerals recognition
