ORIGINAL ARTICLE A multi-matcher system based on knuckle-based features Loris Nanni Æ Alessandra Lumini Received: 2 May 2007 / Accepted: 19 October 2007 / Published online: 6 November 2007 Ó Springer-Verlag London Limited 2007 Abstract We describe a new multi-matcher biometric approach, using knuckle-based features extracted from the middle finger and from the ring finger, with fusion applied at the matching-score level. The features extraction is performed by Radon transform and by Haar wavelet, then these features are transformed by non-linear Fisher trans- form. Finally, the matching process is based on Parzen window classifiers. Moreover, we study a method based on tokenised pseudo-random numbers and user specific knuckle features. The experimental results show the effec- tiveness of the system in terms of equal error rate (EER) (near zero equal error rate). Keywords Biometrics  Knuckle-print verification  Radon transform  Haar wavelet 1 Introduction Among a number of biometric techniques that are used for frequent human identification tasks, systems based on human hands have several advantages over the most common systems based on fingerprints and eye features in spite of they have still not achieved the best security per- formance. First, the features from the human hand are quite stable and relatively easy to be extracted from the hand image. Second the acquisition of the hand biometric is the most acceptable to users probably because it is not asso- ciated with criminal identification. Nevertheless, to our knowledge, only few works (e.g. [5, 9]) study the finger as a biometric characteristic. In [5] it is shown that the lines in the inner skin of knuckle of the finger (named knuckle-print) may be used for iden- tification. In [9] an image-based finger matcher is proposed, where the finger image is projected onto a lower-dimensional space by principal component analysis and then a nearest neighbor classifier is used for the authentication. It should be noted that image-based features and knuckle-based features are partially ‘‘inde- pendent’’ and their complementarity can be exploited by fusion rules. In this work we propose a mono-modal biometric based on knuckle-print. Starting from the hand image, first a localization step is performed to extract the middle finger and the ring finger; then a normalization step is executed to reduce the lighting effects. Different feature extractions, transformations and classifiers have been tested. The best results have been obtained by a multi-matcher based on two different feature extraction methods (Radon transform and the Haar wavelet), followed by a non-linear Fisher transformation [6] for dimensionality reduction and cou- pled with a Parzen window classifier [3]. A fusion at the matching-score level is used and the final decision is based on the ‘‘Sum rule’’. Moreover, we perform experiments considering the combination of the user specific knuckle features with tokenised pseudo-random numbers (BioHashing). The experiments reported in this paper, confirm the results obtained in [7] and show that a multi-matcher system based on the fusion between ‘‘BioHashed’’ features (knuckle features combined with pseudo-random numbers) and solely biometric matcher permits to obtain a near zero equal error rate (EER) also when an ‘‘impostor’’ steals an BioHash key. L. Nanni (&)  A. Lumini DEIS, IEIIT-CNR, Universita ` di Bologna, Viale Risorgimento 2, 40136 Bologna, Italy e-mail: lnanni@deis.unibo.it 123 Neural Comput & Applic (2009) 18:87–91 DOI 10.1007/s00521-007-0160-4