基于稳定性分析的离线手写签名真伪鉴别

发布时间：2018-09-09 10:18

【摘要】：随着21世纪信息技术的飞速发展,中文手写签名的真伪鉴别在很多领域起着重要作用。为了提高离线签名真伪鉴别方法的准确率和鲁棒性,我们提出了一种新的结合签名稳定性分析的混合极限学习机和稀疏表示的层次化分类方案(Integrated Signature Stability and Hierarchical Classifying Method for Signature Verification),简称ISHC-SV。本文需要研究的内容如下:(1)首先对采集的中文手写签名图像进行去噪、细化等预处理操作,之后提取中文签名图像中辨识力度好的静态和伪动态特征集合。(2)针对当前签名鉴别研究均没有考虑稳定性程度导致鲁棒性降低的问题,引入了统计形状模型(Statistical Shape Model,SSM)对签名特征的稳定性进行分析,然后利用方差距离对特征进行量化,最后所有的训练签名被分为稳定的和不稳定的两类。(3)为了解决不同稳定性程度下的签名真伪鉴别分类器选择的问题,利用稳定的签名来训练合适的分类器进行真伪鉴别,同时通过聚类方法将不稳定的签名聚类为模板进行真伪匹配。(4)为了训练高性能的签名真伪鉴别的分类器,提出了一种结合极限学习机和稀疏表示的层次化分类方法以解决签名之间相似度较高的鉴别问题,如熟练仿造的伪签名和真签名。实验结果表明,本文方法与发表在2015年Pattern Recognition期刊中的One-Class SVM方法[43]相比在稳定性较低的签名样本上准确率提高了 3%。此外,本文提出的层次化分类的签名真伪鉴别方法与两种前沿的方法相比总体准确率最高,可以达到95.53%。
[Abstract]:With the rapid development of information technology in the 21st century, the authentication of Chinese handwritten signature plays an important role in many fields. In order to improve the accuracy and robustness of off-line signature authentication, we propose a new hybrid extreme learning machine and sparse representation hierarchical classification scheme, (Integrated Signature Stability and Hierarchical Classifying Method for Signature Verification), for short ISHC-SV., which combines signature stability analysis. The contents of this paper need to be studied as follows: (1) the pre-processing operations such as de-noising and thinning of the collected Chinese handwritten signature images are carried out first. After that, the static and pseudo-dynamic feature sets are extracted from the Chinese signature images. (2) the current signature authentication studies do not consider the problem that the degree of stability leads to the reduction of robustness. In this paper, the statistical shape model (Statistical Shape Model,SSM) is introduced to analyze the stability of signature features, and the variance distance is used to quantify the features. Finally, all the training signatures are divided into two categories: stable and unstable. (3) in order to solve the problem of selecting true and false classifiers under different stability, the stable signatures are used to train the appropriate classifiers for authenticity authentication. At the same time, the unstable signature clustering is used as template to match the authenticity. (4) in order to train the high performance classifier for the authentication of signature authenticity, A hierarchical classification method based on extreme learning machine and sparse representation is proposed to solve the problem of high similarity between signatures, such as fake signature and true signature. The experimental results show that compared with the One-Class SVM method published in the 2015 Pattern Recognition journal, the accuracy of this method is improved by 3% on the less stable signature samples. In addition, compared with the two methods proposed in this paper, the verification method of signature authenticity based on hierarchical classification has the highest overall accuracy and can reach 95.53.
【学位授予单位】：西北大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP309

【相似文献】