Finger vein recognition based on multi-instance

Finger vein is a new member in the biometrics family. Current finger vein recognition algorithms provide high accuracy in good quality images. However, previous researches have shown that the poor quality samples lead to a significant reduction in the accuracy of a single biometric trait system. For non-ideal finger vein images, this paper proposes a method based on multi-instance matching score fusion. First, the features of finger vein images from two or more fingers of a user are extracted by local binary pattern (LBP) algorithm and the matching scores are delivered by Hamming distance. Second, the matching scores are fused using combination-based approach and the decision is made comparing with the threshold. The experimental results show that the proposed method could significantly improve the finger vein recognition performance and the more fingers combined the higher accuracy obtained.