Impact of Writing Order Recovery in Automatic Signature Verification

Abstract

In signature verification, spatio-temporal features offer better performance than the ones extracted from static images. However, estimating spatio-temporal or spatial sequences in static images would be advantageous for recognizers. This paper studies recovered trajectories from skeleton-based images and their impact in automatic signature verification. To this aim, we propose to use a publicly available system for writing order recovery trajectory in offline signatures. Firstly, 8-connected recovered trajectories are generated from our system. Then, we evaluate their impact on the performance of baseline signature verification systems to the original trajectories. Our observations on three databases suggest that verifiers based on distributions are more suitable than those that requiring the exact order of the signatures for the off-2-on challenge.