On-line signature verification based on fusion of global and local information

Automatic signature verification has been an intense research area because of the social and legal acceptance and widespread use of written signatures. It is still a challenging issue because of "small sample size " problem as well as large intra-class variations and, when considering forgeries, small inter-class variations. In order to solve these problems, we propose a two-stage fusion method to get high accuracy. At first, an EDTW (enhanced dynamic time warping) algorithm and a normalized feature measure are proposed to build a classifier based on local features. The former enhances the separability between genuine and forgery signatures, while the latter approaches the problem as a two-class pattern recognition problem, which make it possible to use training signatures as many as possible. However, local method is time and resource consuming, so we then design another classifier based on global features using majority voting rule. We fuse the global and local method by two-sage serial strategy to build an on-line signature verification system. Experimental results on SVC2004 TASK2 show good performance of our system.