Tracking the ballistic trajectory in complex and long handwritten signatures

Abstract

Tracing complex and long handwritten signatures takes an important role in signature verification. Indeed, whether a dynamic signature could be inferred from its static counterpart, improvements would be expected during the automatic verification. An important factor in recovering the tracing of a thinned signature is the feasible and accurate processing of the clusters. A cluster is produced when two or more pieces of handwriting intertwine. Specifically, the challenge is to find which input branch is associated to which output branch and the path between them inside the cluster. In this paper, a novel proposal, based on good continuity criteria derived from both visual perception and movement execution, is developed to solve the paths within the clusters. To this aim, our implementation focuses on a multiscale analysis of the thinned traces and the Dijkstra's algorithm for an effective branch association. Experiments have been carried out with SigComp2009 and SUSIG-Visual signature databases, which are two publicly available Western-based corpus. Promising results have been obtained in our evaluation when studying the success rate in the branches association. The results confirm that processing clusters is important to detect components and a correct cluster branch association improves the writing order recovery performance. Finally, encouraging results have been obtained when performing a global estimation of the writing order in handwriting signatures, in terms of Root Mean Square Error and Dynamic Time Warping.