On-line signature verification using improved segmentation

Abstract

This paper describes on-line signature verification using a HMM (hidden Markov model). The system is based on combining local time-based and shape-based comparison between the input signature and the reference model. We proposed a novel initialization algorithm to avoid the local optimal of the HMM's re-estimation. To get reliable segmentation using the HMM, we adopt a meliorated HMM re-estimation algorithm during training. In addition, we use a novel algorithm to avoid losing important information at the cusps in preprocessing. Tests based on databases and real cases reveal the feasibility of the proposed method.