Quality Aware Online Signature Verification Using Support Vector Machine
Hao Chang · Journal of Information and Computational Science · 2013
A scheme for quality aware online signature verification based on support vector machine was presented and evaluated on the online signature database SVC2004. In this paper, firstly, we conducted experiment on several representative classifiers widely used in online signature verification for comparison. The result showed that DTW and SVM outperformed other classifiers significantly and the EER of SVM was optimistically achieved by 3.67%. Secondly, we divided the signatures into 3 grades with entropy decreasing, that was, incompetent, medium, and excellent according to the quality measure. We found that there were 13 out of 15 Chinese signatures and 16 out of 25 English samples were medium-excellent, which implied that Chinese signature was superior quality and informative. At last, we improved the SVM by adopting soft margin and Gaussian kernel, achieving the EER by 4.18% and 3.67% respectively. Moreover, when quality measure was introduced, the best performance was achieved by 2.48%, which showed 32.43% gain compared with the original Gaussian kernel.