A bilayer image restoration for finger vein recognition

Abstract Finger vein recognition has been widely adopted for human verification because of its high recognition rate and security. Nonetheless, due to light scattering and attenuation in biological tissues, the collected finger vein images are often seriously degraded. This makes finger vein feature representation unreliable, and inevitably impairs the accuracy of finger vein recognition. Exploring effective ways for finger vein image restoration has become an urgent and prevalent topic. In this paper, we analyze the intrinsic factors causing the degradation of finger vein images, and propose a new bilayer restoration method to deal with skin scattering and improve the visibility of finger vein images. Our research is novel in two aspects compared to previous studies. Firstly, an innovative bilayer diffusion model is proposed to precisely describe light scattering in whole finger tissues (including dorsal-side and palm-side tissues). Secondly, we creatively introduce the blur-SURE (Steins unbiased risk estimate) method to yield accurate estimation of the bilayer model’s parameters, and then adopt the multi-Wiener SURE-LET (Linear expansion thresholds) approach to improve the robustness of restoration performance. Experiments performed on two publicly available finger vein databases demonstrate that the proposed method is effective and reliable in finger vein image restoration and enhancement.