Feature-level fusion of hand biometrics for personal verification based on kernel PCA

Abstract

This paper presents a novel method of feature-level fusion (FLF) based on kernel principle component analyze (KPCA). The proposed method is applied to fusion of hand biometrics include palmprint, hand shape and knuckleprint, and we name the new feature as "handmetric". For different kind of samples, polynomial kernel is employed to generate the kernel matrixes that indicate the relationship among them. While fusing these kernel matrixes by fusion operators and extracting principle components, the handmetric feature space is established and nonlinear feature-level fusion projection could be implemented. The experimental results testify that the method is efficient for feature fusion, and could keep more identity information for verification. © Springer-Verlag Berlin Heidelberg 2005.