Fractal indexing in multimodal biometric contexts

Biometric systems aimed at identification and verification range from reliable, slow techniques such as fingerprint matching, to quicker but less reliable techniques such as some face recognition methods. A possible way to improve reliability without losing much on the side of efficiency involves multimodal systems, whose modules process more than one biometric feature. The theory of fractals, having proved itself suitable for effective image indexing methods, can be used for the design of biometric systems too. After introducing the present state of biometrics, fractal based biometric systems are illustrated. A description of unimodal systems is followed by a discussion of multimodal architectures incorporating them. Several issues have to be considered besides the design of the single subsystems: the integration schema, the normalization of results from the single subsystem, the implementation of reliability assessment methods, and the fusion strategy that should be used to integrate the various results into a unified matching score. These issues are discussed in detail, and a fractal based face recognition system is presented: faro. An extension of this system is illustrated, by which faro develops into a multimode system by adding ear recognition. An experimental section tests both unimodal and multimodal faro on several standard databases. The performance is compared to that of other present systems, in order to evaluate the performance enhancement resulting from the multimodal extension. © 2009 Springer-Verlag Berlin Heidelberg.