Abstract: Aiming at the low accuracy, stability and hit rate of classic feature extraction for handwritten digit recognition, a method of feature extraction and classification based on outline architectural feature is proposed. This method repeatedly extracts the outline features of a digit which is converted by some bijective mapping functions, combines the results together by using BP networks, and gets the result through statistical analysis. This feature extraction and classification meth is simple and less time consumed. The experimental results show that this method can improve the hit rate and efficiency of handwritten digit recognition.

Key words: bijective mapping, architectural feature, handwritten digit recognition, Back Propagation（BP） Neural Network

摘要： 针对传统特征提取和分类方法速度慢、稳定性差、识别率低等特点，提出了一种基于外围结构特征提取的手写数字识别方法。该方法多次少量地提取经过双射变换后的图像外围结构特征，对每一次提取的特征结合BP神经网络生成相应的分类器，对不同特征的分类结果进行融合得出手写数字的识别结果。实验结果表明，该特征提取方法实现简单，运算量小，大大提高了脱机手写数字的识别率和效率。

关键词: 双射变换, 结构特征, 手写数字识别, 反向传播神经网络