基于DTW的注意力机制BLSTM在线手写签名认证  

Attentional Mechanism BLSTM Online Handwritten Signature Verification Based on DTW

作　　者：王乐乐 栾方军[1] 师金钢[1] 袁帅[1] WANG Le-le;LUAN Fang-jun;SHI Jin-gang;YUAN Shuai(Information and Control Engineering School,Shenyang Jianzhu University,Shenyang 110168,China)

机构地区：[1]沈阳建筑大学信息与控制工程学院,沈阳110168

出　　处：《小型微型计算机系统》2023年第7期1529-1534,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金面上项目(62073227)资助。

摘　　要：为了提高在线手写签名认证的准确率,设计了一种挖掘签名稳定笔段用于认证的方法.本文提出对签名笔段计算累计差异值矩阵进行匹配;其次采用动态时间规划(DTW)算法计算笔段稳定度;在此基础上,通过以笔段的特征输入双向长短期记忆网络(BLSTM)加注意机制进行处理,从而得到每个用户的稳定签名段集合;最后提取该集合的特征进行分类.该方法在svc2004数据库上进行验证并得到了97.08%的认证率,并在40个用户上取得了1.16%的等误率.该结果表明本文方法能够提高认证精度,并且验证了BLSTM与稳定笔段结合方法的有效性.In order to improveaccuracy of online handwritten signature verification,a method of mining signature stable segments for verification is designed.In this paper,we propose to match cumulative difference matrix of signature segment calculation.Secondly,dynamic time warping(DTW)algorithm was used to calculate segment stability.On this basis,we input features of the segment into the bidirectional Long short-term memory Network(BLSTM)plus the attention mechanism to obtain stable signature segment set of each user.Finally,features of the set are extracted and classified.The method was validated on SVC2004 database and obtained 97.08% authentication rate,and 1.16% equal error rate on 40 users.The results show that the proposed method can improve the authentication accuracy,and verify the effectiveness of BLSTM combined with stable segment.

关 键 词：签名认证 签名分段 双向长短期记忆网络(BLSTM) 注意力机制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]