A new multimodal biometric identification method with small sample size based on iris and palmprint integration

Multimodal biometric identification systems alleviate many problems in unimodal biometric systems, which use a single biometric trait for recognition. We demonstrate that multimodal biometric can play a very important role in one training sample problem. This paper proposed a user-dependent fusion approach, which is based on the investigations that most users have some traits of better class separatability than other traits they have. A new user-dependent fusion algorithm is proposed based on imposter score distribution and fusion binary tree. We then observed that our fusion algorithm improved mean recognition rate by 5.4% on a multimodal biometric database with 120 individuals. It also presents better robustness than other existed fusion methods in all experiments.