HMMs based palmprint identification

Abstract

This paper presents a novel approach of palmprint identification with Hidden Markov Models (HMMs). Palmprint is first aligned and normalized by using the boundary of the fingers. Then the continuous HMMs are used to identify palmprints. The palmprint features are extracted by using Sobel operators and projecting technique. It shows that HMMs with six states and two Gaussian mixtures can obtain the highest identification rate, 97.80%, in one-to-320 matching test. Experimental results demonstrate the feasibility of HMMs on the palmprint identification task. © Springer-Verlag Berlin Heidelberg 2004.