摘要
对手写数字识别技术进行了研究和探讨,提出一种新的多级分类器手写数字识别方法。该识别方法以图像预处理和字符特征提取为基础,采用BP神经网络分类器进行第一级识别、结构特征分类器进行有选择的第二级识别,并提出一种全新的端点特征提取法,大大地简化结构特征分类器的设计。实验结果表明,多级分类器较单一的神经网络分类器的识别率有了明显的提高。
Studies and discusses the technology of handwritten numeral recognition（HNR）, and proposes a new handwritten numeral recognition method based on multi-classifier. This method sets graphics pre-processing and feature extraction as basis and adopts the BP neural network as the first classifier and the structure-feature classifier as the second classifier, which adopts a new method of endpoint-feature extraction and greatly simplifies design of structure-feature classifier. It＇s proved by the testing result that the identification rate of multiclassifier is obviously higher than that of single BP neural network classifier.
出处
《现代计算机》
2009年第2期155-157,共3页
Modern Computer
关键词
手写数字识别
BP神经网络
结构特征分类器
Handwritten Numeral Recognition
BP Neural Network
Structure-Feature Classifier
