Palmprint Recognition Method Based on a New Kernel Sparse Representation Method

摘要

To capture the nonlinear similarity of palmprint image features, a new palmprint recognition method utilizing the kernel trick based sparse representation (KSR) algorithm is proposed in this paper. KSR is in fact an essential sparse coding technique in a high dimensional feature space mapped by implicit mapping function, and it can efficiently reduce the feature quantization error and enhance the sparse coding performance. Here, to reduce the time of sparse coding, the fast sparse coding (FSC) is used in coding stage. FSC solves the L_1 - regularized least squares problem and the L_2 -constrained least squares problem by iterative method, and it has a faster convergence speed than the existing SC model. In test, the PolyU palmprint database used widely in palmprint recognition research is selected. Using the Gauss kernel function and considering different feature dimensions, the task of palmprint recognition obtained by KSR can be successfully implemented. Furthermore, compared our method with general SR and SC under different feature dimensions, the simulation results show further that this method proposed by us is indeed efficient in application.