基于小波包分析和SVM的在线手写签名鉴别  

A Method of On line Handwritten Signature Verification Based on Wavelet Packet Analysis and SVM

作　　者：马海豹[1] 刘漫丹[1] 张岑[1] 

机构地区：[1]华东理工大学自动化研究所,上海200237

出　　处：《华东理工大学学报（自然科学版）》2007年第4期541-545,共5页Journal of East China University of Science and Technology

摘　　要：针对在线手写签名难以提取有效特征的实际情况,提出用小波包分解和单支重构来构造能量特征向量的方法,直接利用各频段成分能量的变化来反映签名的动态特征。给出了衡量各特征识别能力的Fisher准则,并且基于该准则剔除了识别能力差的特征,优化了特征空间。用该方法构造的特征向量能突出反映签名的动态特征。然后采用SVM对签名进行识别。实验证明:采用本文方法识别的正确率高达99.38%,错误拒绝率FRR=0.25%,错误接受率FAR=1.0%,其性能令人满意。It is difficult to extract effective features of on line signature verification, so a new method of dynamic feature extraction is proposed based on decomposition and reconstruction of wavelet packet in this paper. The feature vectors that reflect the energy change of different frequency ranges are constructed by the method and the Fisher criterion was used to select the effective features, With this criterion, some features that contain little discriminating information are discarded in order to simplify feature vector. The feature vectors can reflect the dynamic features of signature effectively, Moreover, SVM is used to recognize signature by the feature vectors constructed above, Experiments show that recognition is correct up to 99.38% , FRR is decreased to 0. 25% and FAR is decreased to 1. 0%. The performance is satisfactory.

关 键 词：手写签名鉴别 小波包分析 特征提取 SVM FISHER准则 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]