Score Selection Techniques for Fingerprint Multi-Modal Biometric Authentication Giorgio Giacinto, Fabio Roli, and Roberto Tronci Department of Electric and Electronic Engineering, University of Cagliari, Piazza D’Armi, I-09123 Cagliari, Italy {giacinto, roli, roberto.tronci}@diee.unica.it Abstract. Fingerprints are one of the most used biometrics for auto- matic personal authentication. Unfortunately, it is often difficult to de- sign fingerprint matchers exhibiting the performances required in real applications. To meet the application requirements, fusion techniques based on multiple matching algorithms, multiple fingerprints, and multi- ple impressions of the same fingerprint, have been investigated. However, no previous work has investigated selection strategies for biometrics. In this paper, a score selection strategy for fingerprint multi-modal authen- tication is proposed. For each authentication task, only one score is dy- namically selected so that the genuine and the impostor users’ scores distributions are mainly separated. Score selection is performed by first estimating the likelihood that the input pattern is an impostor or a gen- uine user. Then, the min score is selected in case of an impostor, while the max score is selected in case of a genuine user. Reported results show that the proposed selection strategy can provide better performances than those of commonly used fusion rules. 1 Introduction Multimodal biometric systems have been proposed to increase the accuracy of authentication systems [4]. According to Prabhakar and Jain [10] multimodal biometric systems can be subdivided into five different scenarios: multiple bio- metrics [1] [5] [11], multiple acquisitions of the same biometry with the same sensor [6], multiple representations, and matching algorithms for the same biom- etry [8] [10], multiple units of the same biometry (for example different fingers) [6], multiple sensors for the same biometry [9]. In this paper we use fingerprints as biometrics. The vast majority of multimodal systems are based on fusion strategies at the score level, so that the score of different matchers are combined to attain a “new” score. Typical fusion rules adopted for multimodal system are the min, max, median, mean, as well as trainable rules such as neural networks [9]. However, the output of different matchers could be exploited not only by fusion rules, but also by dynamic selection mechanisms. In the pattern recognition field, selection mechanisms have been proposed to select, for each input pattern, the classifier that provides the correct output [7]. This formulation of the selection problem