Feature matching in off-line signature verification

The last three decades of research brought many improvements in the field of signature verification. On-line signature verification already has several commercial applications. However, the off-line verification problem is much harder to tackle. One of the tasks is the identification and matching of coherent shapes in different images, which can be a key factor in imporving existing systems. This paper describes an approach to identify, match and compare the different local features (like loops, or baselines) of signatures. This matching allows us not only to evaluate the similarity of the signatures but also to provide a reason for the acceptance or rejection of a signature which is easily interpretable by humans. The method was tested on the corpus of the GPDS Signature Database; experimental results are presented and evaluated. A generalized form of the algorithm could well be applied on other areas of computer vision. Key-Words: signature verification; off-line; classification, feature matching