基于Siamese和BiGRU的在线手写签名验证  

Online handwritten signature verification based on Siamese and BiGRU

作　　者：李相然 郝福珍 LI Xiang-ran;HAO Fu-zhen(Dept.of 7th System,China Electronics Technology Group Corporation,Beijing 100083,China)

机构地区：[1]中国电子科技集团公司第十五研究所系统七部,北京100083

出　　处：《信息技术》2020年第12期91-95,共5页Information Technology

摘　　要：为了解决在线手写签名验证的准确率问题,文中使用深度学习的方法设计了一个基于双向门控循环单元(Bidirectional Gate Recurrent Unit,BiGRU)以及孪生神经网络(Siamese Network)的模型,此方法首先把签名数据的最关键特征提取出来,将它与此签名的参考数据输入到孪生神经网络中,对其进行真伪分类。在文中的实验中,注意力机制(Attention Mechanism)的引入明显提高了此模型的准确性,在SigComp2011手写在线中文数据集中,文中的方法相较此前的最好结果提高了1.33%。In order to solve the problem of accuracy of online handwritten signature verification,a BiGRU(bidirectional gate recurrent unit)and a Siamese network based on deep learning method this paper designs.Firstly,the key features of the signature data are extracted,and then put it and the reference data of the signature into the Siamese network to classify it as being genuine or forged.In this experiment,the usage of attention mechanism significantly improves the accuracy of the model.In the SigComp2011 handwritten online Chinese dataset,the proposed method is 1.33%higher than the previous best results.

关 键 词：在线手写签名验证 深度学习 孪生神经网络 双向门控循环单元 注意力机制 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]