基于方向线素特征的孟加拉手写数字识别  

Handwritten Bangla Numerals Recognition Based on Directional Element Feature

作　　者：林颖[1] 吕岳[1] 

机构地区：[1]华东师范大学计算机科学与技术系,上海200241

出　　处：《计算机工程》2009年第15期185-186,189,共3页Computer Engineering

基　　金：教育部新世纪优秀人才支持计划基金资助项目(NCET-05-0430)

摘　　要：根据孟加拉数字的特点,将方向线素特征应用于孟加拉手写数字识别的特征提取,并辅以端点和交叉点特征,采用BP神经网络作分类器进行识别。利用从实际孟加拉信封图像中采集到的手写体数字作为样本进行实验,结果表明,该方法的识别率和可靠性分别达到97.63%和98.77%。This paper presents a novel approach to automatic recognition of handwritten Bangla numerals. An effective method for Bangla numerals＇ feature extraction is carried out by using Directional Element Feature（DEF） combined with weighted Ending and Crossing Point（ECP） features. BP neural network is employed as a classifier. Experimental result on the numeral images obtained from real Bangladesh envelopes show that the proposed recognition method achieves a satisfied result on both recognition and reliability with the rates of 97.63% and 98.77% respectively.

关 键 词：方向线素特征 孟加拉手写数字 BP神经网络 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]