Algorithm for retrieval and verification of personal identity using bimodal biometrics

Abstract A scheme for integrating two modalities for person recognition is proposed. The system works in the open-set identification mode, which means that it identifies with possibility of rejecting impostors. One classifier is used to retrieve a ranking of n most probable labels, which are consecutively verified using the second classifier. Verification procedure takes into consideration the probability that a label is genuine. The system has been implemented to combine face identification and speaker verification. Testing demonstrated that fusion yields up to 3.9% reduction of FRR at constant FAR with 67% increase of system response time.