基于混合特征的孟加拉手写体数字识别  

Bangla handwritten numeral recognition based on blend features

作　　者：刘春丽[1] 吕淑静[1] 

机构地区：[1]上海工商外国语学院,上海201300

出　　处：《计算机工程与应用》2007年第20期214-215,共2页Computer Engineering and Applications

摘　　要：根据孟加拉手写体数字的特点,用Kirsch算子提取象素的水平、垂直、右对角线以及左对角线的特征矢量,并与字符图像的密度特征相结合,采用BP算法训练的MLP网络作分类器进行识别。最后,用从实际孟加拉信封图像中采集到的手写体数字作样本进行实验,达到了96.1%的识别率。A recognition system based on blend features is proposed for Bangla handwritten numeral.The direction features,including horizontal vector,vertical vector,right diagonal vector and left diagonal vector,which combine with density features,are extracted by Kirsch operator.Then the numerals are recognized by means of BP network.Experimental results show that the proposed method obtains 96.1% correct recognition rate.

关 键 词：特征提取 KIRSCH算子 孟加拉手写体数字 BP神经网络 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]