基于细粒度动态特征的摹仿签名书写人识别  

Identification of Imitative Signature Write rs Based on Fine-grained Dynamic Features

作　　者：齐明乐 池长江 李毅峰 申思[1] QI Ming-le;CHI Chang-jiang;LI Yi-feng;SHEN Si(Zhejiang Police College,Hangzhou Zhejiang310053,China)

机构地区：[1]浙江警察学院,浙江杭州310053

出　　处：《计算机仿真》2024年第7期263-268,共6页Computer Simulation

摘　　要：电子签名笔迹逐渐取代传统笔迹,电子签名的真伪鉴别成为公安、司法鉴定领域的难题。于是提出了细粒度电子签名笔迹动态特征提取方法,利用K近邻、决策树、随机森林、支持向量机等监督学习方法综合分析摹仿电子签名的动、静态特征,建立摹仿电子签名笔迹书写人识别模型。实验结果显示,基于K近邻算法的书写人识别模型表现最好,正确率0.917,精确率0.906,召回率为0.871,AUC为0.965。实验表明,笔迹动态特征能够显著提升摹仿签名书写人识别模型性能,增加样本类别数或者减低样本数量均会降低模型的识别能力。Electronic signature gradually rep laces traditional handwriting,and the authe ntication of electronic signature has become a difficult problem in the field of public security and judicial authen tication.In this paper,a finegrained electronic signature handwriting dynamic feature extraction method was proposed,and sup ervised learning methods including K-nearest neighbor,decisi on tree,random forest and support vector mac hine were used to comprehensively analyze the dynamic features and static features of the imitation electronic signature.A classification model of the imitation electronic signature was established.The experiment results demonstrate that the writer recognition model based on K-nearest neighbor algorithm has the best performance,whose accuracy rate,precision rate,recall rate and the AUCis 0.917,0.906,0.871,and0.965,respectively.The experiment shows that the dynamic features of electronic handwriting can significantly improve the performance of writer classificati on model,the model's recognition ability declines when the categories of training sample or the number of training set decreasing.

关 键 词：电子签名笔迹 动态笔迹特征 机器学习 摹仿签名 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]