Neurocomputing 69 (2006) 1706–1710 Letters An advanced multi-modal method for human authentication featuring biometrics data and tokenised random numbers Alessandra Lumini, Loris Nanni à DEIS, IEIIT—CNR, Alma Mater Studiorum, Universita` di Bologna, Viale Risorgimento 2, 40136 Bologna, Italy Received 7 September 2005; received in revised form 23 January 2006; accepted 24 January 2006 Communicated by R.W. Newcomb Abstract In this work, we propose a multi-modal method that combines the scores of selected fingerprint matchers with the scores obtained by a Face Authenticator where the facial features are combined with pseudo-random numbers. We propose a novel method to combine the scores of fingerprint matchers based on random subspace Ensemble and we test the method on the systems submitted to FVC2004. Moreover, we show that methods based on tokenised pseudo-random numbers and user specific biometric features are highly dependent upon a parameter, the hashing threshold; we demonstrate that using an ensemble of classifiers it is possible to solve this problem leading to a considerable performance improvement. Finally, we study the fusion among the scores obtained by a Face Authenticator (where the face features are combined with pseudo-random numbers) and the scores of the systems submitted to FVC2004. r 2006 Elsevier B.V. All rights reserved. Keywords: Fingerprint; Multimodal fusion; Pseudo-random numbers 1. Introduction The increasing interest in a wide variety of practical applications for automatic personal identification and authentication has resulted in the popularity of biometric recognition systems [8]. As a consequence, recent efforts have been conducted in order to establish common evaluation scenarios enabling a fair comparison between competing systems. For instance, in the field of fingerprint recognition, a series of International Fingerprint Verifica- tion Competitions (FVC) [7] have received great attention both from the academy and the industry. Denial of access in biometric systems greatly impacts on the usability of the system by failing to identify a genuine user. Multimodal biometrics can reduce the probability of denial of access without sacrificing the false acceptation performance, the key is the combination of the various biometric character- istics at the feature extraction, match score, or decision level [8]. In [4], the authors, in order to solve the problem of high false rejection, proposed a novel two-factor authenti- cator based on iterated inner products between tokenised pseudo-random numbers and the user fingerprint feature; in this way a set of user compact codes can be produced which is named ‘‘BioHashing’’. The possible drawback of BioHashing is the low performance when an ‘‘impostor’’ B steals the pseudo-random numbers (hash key) of A and tries to authenticate as A. When this problem occurs, the performance of BioHashing can be lower than that obtained using only the biometric data. In [6], the authors proposed a multimodal system based on the fusion between ‘‘BioHashed’’ face features (face features combined with pseudo-random numbers) and the scores obtained by some of the fingerprint verification methods submitted to FVC2004. They showed that the fusion permits to obtain good performance (similar to that obtained by the standard fusion between face and fingerprint) also when an ‘‘impostor’’ steals an hash key. In this paper, we improve that result by proposing to use a random subspace (RS) method [3] to combine the scores of fingerprint matchers. We show that a fusion of classifiers based on RS permits to obtain an equal error ARTICLE IN PRESS www.elsevier.com/locate/neucom 0925-2312/$ - see front matter r 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.neucom.2006.01.010 à Corresponding author. Tel.: +39 349 3511673; fax: +39 547 338890. E-mail address: lnanni@deis.unibo.it (L. Nanni).