A distance-based approach to modelling reference signature for verification

Abstract

Signature verification is very important in verification and authentication systems due to its widespread use in our daily lives. Previous research in this field has mainly focused on optimizing features and matching algorithms. In this paper, we present an effective alternative approach for producing a reference signature that generalizes the properties of the given sample signatures. Two experiments are conducted: using the traditional Average Reference Signature (ARS) and using the new Average Permutative Difference Signature (APDS) approach. The Discrete Fourier Transform is applied on signature information and the Fast Dynamic Time Warping algorithm is used as a matching algorithm. Experiments were carried out on the skilled forgeries of the SVC2004 database, yielding False Rejection Rates of 0.31 and 0.51; and False Acceptance Rates of 1.84 and 1.36, for the two approaches (ARS and APDS) respectively.