IEEE Trans Cybern. 2018 Feb;48(2):611-624. doi: 10.1109/TCYB.2017.2647826. Epub 2017 Jan 17.
This paper explores the utility of information derived from the dynamic time warping (DTW) cost matrix for the problem of online signature verification. The prior works in literature primarily utilize only the DTW scores to authenticate a test signature. To the best of our knowledge, the characteristics of the warping path (used for the alignment) in the cost matrix is hardly exploited for verification of online signatures. Accordingly, we devise a score that utilizes the information from the cost matrix and warping path alignments. We subsequently consider its fusion (using a sum rule combiner) with the DTW score for authenticating the veracity of a test signature. In addition, a minor modification is suggested with regards to the set of features employed for matching the signatures. We introduce a spacing parameter for feature extraction and demonstrate its applicability in increasing the separation between the distribution of genuine and forgery signatures for an user. Our method has been tested on two publicly available online signature databases namely the SVC-2004 Task 2 and MCYT-100. We report reduction in error rates over the traditional DTW framework.
本文探讨了从动态时间规整 (DTW) 代价矩阵中提取信息在在线签名验证问题中的应用。文献中的先前工作主要仅使用 DTW 得分来验证测试签名。据我们所知，代价矩阵中的弯曲路径（用于对齐）的特征在验证在线签名时几乎没有被利用。因此，我们设计了一个利用代价矩阵和弯曲路径对齐信息的得分。随后，我们考虑将其与 DTW 得分融合（使用和规则组合器）来验证测试签名的真实性。此外，还针对用于匹配签名的特征集提出了一个小的修改。我们引入了一个特征提取的间距参数，并证明了它在增加用户真实和伪造签名分布之间的分离方面的适用性。我们的方法已经在两个公开可用的在线签名数据库 SVC-2004 任务 2 和 MCYT-100 上进行了测试。我们报告了在传统 DTW 框架上错误率的降低。
