A novel normalization technique for multimodal biometric systems

Abstract

A multimodal biometric system consolidates multiple biometric sources and mitigates the limitations of the unimodal biometric system. The consolidation of information can be done at various levels of fusion. In this paper, a normalization technique for score-level fusion based on a new anchor, which is computed from the raw score set, has been proposed. This new anchor is independent of the statistatical properies of the biometric system (e.g., equal error rate). This work focuses on the improvement of the multimodal biometric system that consists of at least one weak classifier and does not have a prior knowledge of the genuine/impostor score distribution. The experimental results show that the performance of simple non-weighted fusion preceded by our EER independent anchor based normalization technique is better than that of the weak classifier of the system and is superior to that of two other normalization methods.