Computer Engineering and Applications ›› 2007, Vol. 43 ›› Issue (20): 214-215.

• 工程与应用 • Previous Articles     Next Articles

Bangla handwritten numeral recognition based on blend features

LIU Chun-li,LV Shu-jing   

  1. Shanghai Industry & Commerce Foreign Language College,Shanghai 201300,China
  • Received:1900-01-01 Revised:1900-01-01 Online:2007-07-11 Published:2007-07-11
  • Contact: LIU Chun-li

基于混合特征的孟加拉手写体数字识别

刘春丽,吕淑静   

  1. 上海工商外国语学院,上海 201300
  • 通讯作者: 刘春丽

Abstract: 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.

Key words: feature extraction, Kirsch operator, Bangla handwritten numeral, BP network

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。

关键词: 特征提取, Kirsch算子, 孟加拉手写体数字, BP神经网络