Confidence Partition and Hybrid Fusion in Multimodal Biometric Verification System

Sum rule fusion is a very promising multimodal biometrics fusion approach. However, it is proposed not to widely applying it across the multimodal biometrics, score space. By examining the score distributions of each biometric matcher, it can be seen that there exist confidence regions which enable the introduction of the Confidence Partition in multimodal biometric score space. It is proposed that the Sum rule can be replaced by the Min or the Max rule in the Confidence Partition to further increase the overall verification performance. The proposed idea which is to apply the fusion rules in a hybrid manner has been tested on two publicly available databases and the experimental results shows 0.3% similar to 2.3% genuine accept rate improvement at relatively low false accept rate.