Source: Recherche Informatique Vietnam & Francophonie
Authors: B. Ly-Van, R. Blouet, S. Renouard, S. Garcia-Salicetti, B. Dorizzi, G. Chollet
Keywords: Additive Tree Classifier, Dynamic Time Warping, Fusion, Hidden Markov Models, Gaussian Mixture Models, Multimodal Identity Verification, Support Vector Machines
Abstract: This article presents the main idea of a Biometric Authentication System (BAS). We will briefly describe BIOMET database, constructed for this purpose and composed of five modalities: face, hand-shape, fingerprint, online signature and speech. We have used two user-friendly biometric modalities of this database to construct the unimodal verification systems based on online signature and speech. The Signature Verification system relies on Hidden Markov Models (HMMs), and two kinds of Speaker Verification systems have been designed. The first one is text-dependent, using Dynamic Time Warping (DTW) to compute a decision score. The second one is text-independent and use Gaussian Mixture Models (GMMs) to compute the likelihood ratio of an utterance. We present also two classical learning-based fusion techniques: an additive CART-trees classifier built with boosting and Support Vector Machines (SVMs). In particular, the signature modality was fused with clean and noisy speech, at two different levels of degradation. We have at our disposal 68 real persons, divided into two equal bases BAF and BTF to train and test the unimodal systems as well as the fusion systems. The performances of the systems are evaluated on 20 different couples (BAF, BTF), randomly chosen from 68 original real persons. In the non noisy case, we have dropped down the total error rate to 2.21% from 7.65% for the best unimodal system.