计算机工程与应用 ›› 2018, Vol. 54 ›› Issue (8): 112-118.DOI: 10.3778/j.issn.1002-8331.1611-0238

• 模式识别与人工智能 • 上一篇    下一篇

层次化分类的离线中文签名真伪鉴别方法

魏佳敏1,冯  筠1,卜起荣1,高  原2,赵  妍1   

  1. 1.西北大学 信息科学与技术学院,西安 710127
    2.西北大学 经济管理学院,西安 710127
  • 出版日期:2018-04-15 发布日期:2018-05-02

Off-line Chinese handwriting signature verification with hierarchical classification

WEI Jiamin1, FENG Jun1, BU Qirong1, GAO Yuan2, ZHAO Yan1   

  1. 1.College of Information Science and Technology, Northwest University, Xi’an 710127, China
    2.College of Economics and Management, Northwest University, Xi’an 710127, China
  • Online:2018-04-15 Published:2018-05-02

摘要: 为了改进中文手写签名真伪鉴别系统的性能,提出了一种混合极限学习机和稀疏表示的层次化分类方法。首先,利用极限学习机强大的泛化能力和鲁棒性,对较易识别的伪签名进行分类,如随机伪造的签名;接着,利用稀疏表示分类具有的精准描述性能,设计签名数据字典,对较难识别的伪签名进行分类,如熟练伪造的签名。实验结果表明,层次化分类的签名鉴别方法与前沿的两种方法相比总体准确率最高,达到了95.53%。

关键词: 签名真伪鉴别, 层次化分类, 极限学习机, 稀疏表示分类, 静态特征, 伪动态特征

Abstract: This paper proposes a hierarchical classification method for Chinese handwriting signature verification called HCSV(Hierarchical Classifier for Signature Verification, HCSV), which combines Extreme Learning Machine(ELM) with Sparse Representation(SR). First of all, taking advantage of the generalization ability and robustness of ELM, random forged signatures which can easily be identified are distinguished. Afterwards, the premeditated forged signatures are selected from the real signatures by sparse representation classifiers which have good ability of feature description. Experimental results show that the proposed method achieves 95.53% verification accuracy, which is better than two state-of-art methods.

Key words: signature verification, hierarchical classifying, extreme learning machine, sparse representations classifier, static features, pseudo-dynamic features