On-line signature verification using model-guided segmentation and discriminative feature selection for skilled forgeries

Abstract

The paper describes an online signature verification system using model-guided segmentation and discriminative feature selection for skilled forgeries. The system is based on segment-to-segment comparison between the input signature and the reference model. To obtain a consistent segmentation, we propose a model-guided segmentation, which segments an input signature by the correspondence with the reference model. To reject skilled forgeries effectively, we use a discriminative feature selection. It is motivated from the observation that a skilled forger can imitate the shape of the genuine signature better than even the owner, that is some features distinguish skilled forgeries from genuine signatures, though some features distinguish only random forgeries. For random forgeries and skilled forgeries respectively, we select the discriminative features among all the features according to the distance between references and forgeries. In the experiment, we collected 1000 genuine signatures and 1000 skilled forgeries. The result showed that the proposed method gave more stable segmentation, and the discriminative feature selection eliminated about 62% of the errors.