Quality controlled multimodal fusion of biometric experts

The quality of biometric samples used by multimodal biometric experts to produce matching scores has a significant impact on their fusion. We address the problem of quality controlled fusion of multiple biometric experts and focus on the fusion problem in a scenario where biometric trait quality expressed in terms of quality measures can be coarsely quantised. We develop a fusion methodology based on fixed rules that exploit the respective advantages of the sum and product rules and can be easily trained. We show in experimental studies on the XM2VTS database that the proposed method is very promising.