A supervised method to discriminate between impostors and genuine in biometry Loris Nanni * , Alessandra Lumini DEIS, University of Bologna, via Venezia 52, 47023 Cesena, Italy article info Keywords: Score normalisation Supervised classification Unconstrained cohort normalisation Biometric identification abstract In this paper, we describe a supervised technique that allows to develop a more robust biometric system with respect to those based directly on the similarities of the biometric matchers or on the similarities normalised by the unconstrained cohort normalisation. In order to discriminate between genuine and impostors a quadratic discriminant classifier is trained using four features: the similarities of the biometric matcher; the similarities of the biometric matcher after the unconstrained cohort normalisation (UCN); the average scores among the test pattern and the users that belong to the background model; the difference between the user-specific threshold and the user-independent threshold. The proposed technique is validated by extensive experiments carried out on several biometric data- sets (palm, finger, 2D and 3D faces, and ear). The experimental results demonstrate that the capabilities provided by our supervised method can significantly improve the performance of a standard biometric matcher or the performance of the standard UCN. Ó 2009 Elsevier Ltd. All rights reserved. 1. Introduction One of the important problems in biometrics (Jain, Ross, & Pra- bhakar, 2004) is the score normalisation. The concept of score nor- malisation was originally introduced for the speaker recognition, due to the fact that the statistical speaker matchers provide the ver- ification score as the probability of the observed test utterance x, gi- ven the target model k (Ariyaeeinia, Sivakumaran, Pawlewski, & Loomes, 1999; Fortuna, Sivakumaran, Ariyaeeinia, & Malegaonkar, 2004). In the literature (Alsaade, Ariyaeeinia, Malegaonkar, Pawlew- ski, & Pillay, 2007) it is shown that score normalisation can improve the performance of other biometric matchers, and in particular the unconstrained cohort normalisation (UCN) can be very useful to sep- arate the genuine scores from the impostors scores. The application of UCN can improve the biometrics performance since, if an ade- quately large set of background models is available, an impostor tar- geting a particular client model is likely to match some of the background models more strongly (Alsaade et al., 2007). Moreover, it has been shown (Ariyaeeinia et al., 1999) that the UCN approach works effectively regardless of whether the operating framework is probabilistic or non-probabilistic. To the best of our knowledge the UCN was applied only in speaker recognition and face recognition (Alsaade et al., 2007). In this paper we extend the application of UCN to four different bio- metric characteristics: palm, finger, face (both 2D and 3D) and ear, discovering that UCN gives a considerable advantage in distin- guishing genuine users from impostors only in some of the tested biometrics. In particular our results demonstrate that UCN is well suited for ear recognition, while it is not suited for palm and finger recognition. Moreover, in this paper we propose a supervised method to dis- criminate between impostors and genuine users based on a qua- dratic discriminant classifier trained using the following four features: the similarities of the biometric matcher; the similarities of the biometric matcher after the unconstrained cohort normali- sation (UCN); the average score among the test pattern and the users that belong to the background model; the difference between the user-specific threshold and the user-independent threshold. Our tests, carried out on five well-known biometric benchmarks show that our supervised method improves not only the perfor- mance obtained using the scores provided by a biometric matcher but also the performance obtained using standard UCN. The valid- ity of our approach is demonstrated considering many different indicators (EER, EUC, DET-curve and ROC curve) at the state-of- the-art of biometric performance evaluation. The paper is organised as follows: in Section 2 the new super- vised technique for biometric verification is explained, in Section 3 the experimental results are presented and commented. Finally, in Section 4 some concluding remarks are given. 2. System description In this section, we describe a novel supervised technique that permits to better discriminate between impostors and genuine 0957-4174/$ - see front matter Ó 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2009.01.037 * Corresponding author. E-mail addresses: loris.nanni@unibo.it (L. Nanni), alessandra.lumini@unibo.it (A. Lumini). Expert Systems with Applications 36 (2009) 10401–10407 Contents lists available at ScienceDirect Expert Systems with Applications journal homepage: www.elsevier.com/locate/eswa