摘要
该文分析了手写签名样本的特征值在特征空间上的分布。在此基础上,直接从神经元分类功能的物理意义出发,设计了具有非线性边界的,用于手写签名认证的神经网络分类器,妥善地解决了实际应用中,由于真实签名样本数量少和伪签名样本缺乏,不能训练神经网络的问题,取得了较好的认证结果。
In this paper the distribution of the feature vectors of signatures is analyzed.In light of the distribution's characteristics,a simple nonlinear neural network classifier is designed.The neural network's connection weights are de-termined directly from the physical meaning of the classifying edge,so the problem caused by the lack of training sam-ples is avoid.This classifier is applied to the signature verification,and a good correct rate is obtained.
出处
《计算机工程与应用》
CSCD
北大核心
2002年第19期88-89,92,共3页
Computer Engineering and Applications