Online signature verification based on legendre series representation. Consistency analysis of different feature combinations

Abstract

In this paper, orthogonal polynomial series are used to approximate the time functions associated to the signatures and the coefficients in these series are used as features to model them. A novel consistency factor is proposed to quantify the discriminative power of different combinations of time functions related to the signing process. Pen coordinates, incremental variation of pen coordinates and pen pressure are analyzed for two different signature styles, namely, Western signatures and Chinese signatures from a publicly available Signature Database. Two state-of-the-art classifiers, namely, Support Vector Machines and Random Forests are used in the verification experiments. The obtained error rates are comparable to results reported over the same signature datasets in a recent signature verification competition. © 2012 Springer-Verlag.