Finger-Knuckle-Print recognition performance improvement via multi-instance fusion at the score level

Abstract Fusion of multiple instances within a modality for improving the performance of biometric verification has attracted much attention in recent years. In this letter, we present an efficient Finger-Knuckle-Print (FKP) recognition algorithm based on multi-instance fusion, which combines the left index/middle and right index/middle fingers of an individual at the matching score level. Before fusing, a novel normalization strategy is applied on each score and a fused score is generated for the final decision by summing the normalized scores. The experimental results on Poly-U FKP database show that the proposed method has an obvious performance improvement compared with the single-instance method and different normalization strategies.