Spiral Based Run-Length Features for Offline Signature Verification

Abstract

Automatic signature verification is one of the main modes to verify the identity of the individuals. Among the strategies to describe the signature in the verifiers, run-length features have attracted the attention of many researchers. This work aims to upgrade the classical run-length distribution as an additional representation for off-line signatures. Specifically, we add a fifth direction to the four classical directions of run-length features. Such fifth direction runs the signature in a spiral way providing an outside to inside view of the signature. This paper evaluates the performance of the new run-length direction combined with the classical ones. For classification purposes, we used a one-class support vector machine. Additionally, we study how to combine the new direction with the previous four original ones at both feature and score levels. Our results validate the use of this novel direction in run-length features in our own experiments and external international competition in signature verification.