Contactless palm vein identification using multiple representations

This paper investigates some promising approaches for the automated personal identification using contactless palmvein imaging. We firstly present two new palmvein representations, using Hessian phase information from the enhanced vascular patterns in the normalized images and secondly from the orientation encoding of palmvein line-like patterns using localized Radon transform. The comparison and combination of these two palmvein feature representations, along with others in the palmvein literature, is presented for the contactless palmvein identification. We also evaluate the performance from various palmvein representations when the numbers of training samples are varied from minimum. Our experimental results suggest that the proposed representation using localized Radon transform achieves better or similar performance than other alternatives while offering significant computational advantage for online applications. The proposed approach is rigorously evaluated on the CASIA database (100 subjects) and achieves the best equal error rate of 0.28%. Finally, we propose a score level combination strategy to combine the multiple palmvein representations. We achieve consistent improvement in the performance, both from the authentication and recognition experiments, which illustrates the robustness of the proposed schemes.