Online Signature Verification Based on Dynamic Feature Segmentation and 3-Step Matching

Abstract

We propose a new on-line signature verification system based on dynamic feature segmentation and 3 step matching. Conventional segmentation methods are based on the shape of an input signature and it can be forged easily. Since our segmentation method is based on dynamic features such as speed and pressure of a pen, it makes a signature difficult to forge. Then the segments are associated with those of model signatures using augmented dynamic programming (DP) which exploits static features as a restriction condition in order to increase the reliability of matching between two segments. Also whole matching procedure is composed of three steps to minimize two types of errors, Type I and Type II. Our method is very useful to discern a forgery from input signatures. Experiments show the comparing results among on-line signature features, the basis of weights decision for each feature, and the validity of segmentation based on dynamic feature points.