Feature Weighted Support Vector Machines for Writer-Independent On-Line Signature Verification

In this paper we present a novel framework for writer-independent on-line signature verification. This framework utilises a dynamic time warping-based dichotomy transformation and a writer-specific dissimilarity normalisation technique, in order to obtain a robust writer-independent signature representation in dissimilarity space. Support vector machines are utilised for signature modelling and verification. Linear and radial basis function kernels are investigated. In the case of the radial basis function kernel, both conventional and feature weighted variants are considered. We show that the non-linear kernel significantly outperforms its linear counterpart. We also show that the incorporation of feature weights into the non-linear kernel function consistently improves verification proficiency. When evaluated on the Philips signature database, which contains 1530 genuine signatures and 3000 amateur skilled forgeries from 51 writers, we show that equal error rates of 1.26% and 3.52% are expected when 15 and 5 genuine reference samples are considered per writer. This performance estimate compares favourably with those of existing systems also evaluated on this data set. Furthermore, there is sufficient evidence to suggest that further investigation into the feature set considered, as well as the feature weighting strategy utilised, may further improve performance.