Online Signature Verification: Improving Performance through Pre-classification Based on Global Features

Abstract

In this paper, a pre-classification stage based on global features is incorporated to an online signature verification system for the purposes of improving its performance. The pre-classifier makes use of the discriminative power of some global features to discard (by declaring them as forgeries) those signatures for which the associated global feature is far away from its respective mean. For the remaining signatures, features based on a wavelet approximation of the time functions associated with the signing process, are extracted, and a Random Forest based classification is performed. The experimental results show that the proposed pre-classification approach, when based on the apppropriate global feature, is capable of getting error rate improvements with respect to the case where no pre-classification is performed. The approach also has the advantages of simplifying and speeding up the verification process.