A. Campilho and M. Kamel (Eds.): ICIAR 2008, LNCS 5112, pp. 1033–1040, 2008. © Springer-Verlag Berlin Heidelberg 2008 A New Data Normalization Function for Multibiometric Contexts: A Case Study Maria De Marsico 1 and Daniel Riccio 2 1 Dipartimento di Informatica, Università degli Studi di Roma "La Sapienza" demarsico@di.uniroma1.it 2 Dipartimento di Matematica e Informatica, Università di Salerno, Via Ponte don Melillo, 84084, Fisciano (SA) driccio@unisa.it Abstract. It has been not possible yet to identify a physical or behavioural fea- ture able by itself to identify a person in a way satisfying the acceptability and reliability constraints imposed by real applications. As a consequence the present trend is towards multimodal systems. Data normalization problem is crucial when fusing results from different subsystems. We introduce a new nor- malization function, the mapping function, able to overcome the limitations of commonly used techniques. In this work we also test it on a real hierarchical system obtained by the novel combination schema of the three different bio- metries face, ear and fingerprint. Experimental results in the final part of our work provide a positive feedback about assertions within the body of the paper. Keywords: biometrics, score normalization, multimodal systems. 1 Introduction Biometrics allows recognizing an individual based on physical or behavioural features and its use potential spans from authentication and access control, to identification of missing or dangerous persons. Many biometric techniques exist nowadays, including those based on fingerprints, hand conformation, iris scanning, features from face or ears, voice or handwriting. Most of the biometric technologies [5] developed so far are intrusive: as an example, the subject might need to put a finger or the hand on a sensor. However, passive technologies also exist, requiring little or no activity from the subject: examples are face, ear and voice recognition. Promising though it is, the biometric approach is in most cases still limited to controlled or experimental settings. In any case, present systems generally rely on a single biometry: this makes them vulnerable to possible attacks, and little robust with respect to a number of problems. Examples are acquisition errors, or the possible non universality of a biometric fea- ture. A multi-biometric system provides an effective solution for such problems, as flaws of an individual system can be compensated by alternative biometries [9]. A normalization process is crucial for multi-biometric systems, in order to consistently fuse data from different modules [6]. In this work we propose a hierarchical model, where each subject in the output list of the first level subsystems (face and ear) is