基于BP神经网络的脱机手写数字识别  

Off-line Handwritten Numeral Recognition Using Topological Feature Structures

作　　者：杨金伟[1] 段会川 

机构地区：[1]山东师范大学信息科学与工程学院,济南250014 [2]不详

出　　处：《信息技术与信息化》2008年第4期49-50,54,共3页Information Technology and Informatization

摘　　要：脱机手写体数字识别有着重大的使用价值,特征提取占据了重要的位置,本文针对手写体数字识别中单一识别方法的局限性,提出采用BP神经网络进行识别,并且提出了一种特征提取方法。采用BP神经网络,利用其良好的监督学习功能进行识别,结合提取的降维数字符号的特征,能较好的识别出手写数学符号。BP神经网络(Back-Propagation),又称误差反向传递神经网络,是一种依靠反馈值来不断调整节点之间的连接权值而构建的一种网络模型。最后,在Mnist手写数据库上的试验结果表明,该方法具有较好的识别率和较高的可靠性。Off- line handwritten numeral recognition has great value, in which feature extraction occupy an important position. This paper uses BP neural network to recognize handwritten numerals against the limitations of a single recognition method, and put forward a method of feature extraction. BP network has good learning supervision power. With the symbolic features of the extracted dimension - fallen figures, the handwriting numerals can be well recognized. BP neural network （Back -Propagation） , also known as the reverse error transferred neural network, is a network model of a feedback on the value of continuous adjustment to connected value among the nodes. Finally, the test on the Mnist handwriting database shows that the method has a good recognition rate and high reliability.

关 键 词：预处理 BP神经网络 手写数字识别 

分 类 号：TP391.43[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]