An Effective Feature Selection Method for On-line Signature based Authentication

Abstract

In this paper, we tackle the problem of identifying the relevant set of features that helps achieving accurate on-line signature based authentication. There exists a large set of features that can be acquired from the original signal or derived from it. Taking into account the whole set of features in the authentication process is time consuming. Furthermore, not all features are relevant and some of them are redundant. Consequently, finding the minimal set of relevant features is a prerequisite to perform fast authentication while achieving better accuracy. This feature selection task is combinatorial in nature. In our work, we handle it using a Discrete Quantum behaved Particle Swarm Optimization strategy (DQPSO). The space of possible feature sets is explored according to a QPSO dynamic where each set is encoded in terms of a binary representation. Data sets from SVC 2004 data base have been used in our experiments. Very encouraging results have been obtained.