基于背景域串行分类器的手写字符识别算法研究  

Research on Handwritten Character Recognition Algorithm Based on Serial Classifier in Background Domain

作　　者：罗佳[1] 刘大刚[2] 杨姝[1] 杨菊英[1] LUO Jia;LIU Da-gang;YANG Shu;YANG Ju-ying(Computer Department,ChengDu College Of University Of Electronic Science And Technology of China,Changdu 610000,China;National Key Laboratory of Southwest Jiaotong University,Chengdu 610031,China)

机构地区：[1]电子科技大学成都学院计算机系,四川成都610000 [2]西南交通大学国家重点实验室,四川成都610031

出　　处：《微电子学与计算机》2018年第10期41-45,共5页Microelectronics & Computer

基　　金：国家自然科学基金(项目编号:51208434);中央高校基本科研业务费专项资金(项目编号:SWJTU12CX070)

摘　　要：本文提出一种基于背景域串行分类器的手写字符识别算法.该算法可对52种手写大小写字母和10种数字,共62种字符进行通用识别,主要包括两大模块:一是提取字符图像背景域特征;二是采用MQDF和SVM两级分类器串行组合.首先MQDF分类器通过背景域的凹凸特征进行识别初期的粗分类,然后SVM分类器通过背景域的宽高比和横纵向交截特征进行识别后期的细分类.实验结果表明,本文算法不仅通用性较好,而且识别率、拒识率以及识别速度都优于同类算法.This paper proposed algorithm of handwritten character recognition based on serial classifier domain background. The algorithm can be used for 52 handwritten letters and 10 kinds of numbers and a total of 62 characters. It mainly includes two major modules： the first is to extract the character image background domain features; The seconde is the serial combination of MQDF and SVM two level classifiers. First of all, the MQDF classifier is rough classification at the initial stage of recognition through the concave and convex features of the background domain. Then the SVM classifier is classified by the width to height ratio of the background domain and the cross section feature to identify the fine classification later. The experimental results show that the algorithm in this paper is not only good in generality, and the recognition rate, the rejection rate and the recognition speed are superior to those of the same algorithm.

关 键 词：背景域 凹凸特征 MQDF SVM 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]