FVHOC: An Efficient Finger Vein Recognition Based on Histogram of Oriented Curvatures

Abstract

Finger-vein recognition has become a promising modality in biometric authentication. However, the traditional template-based and handcrafted methods are limited in representing the geometric properties of vein ridges, such as curvature continuity and orientation coherence, making them sensitive to low contrast, misalignment, and image quality variations. To bridge this gap, this article proposes FVHOC, an efficient finger-vein recognition method that explicitly leverages second-order structural information of vein ridges. By integrating Hessian-based curvature extraction with blockwise histogram encoding, FVHOC encodes the second-order structural characteristics of finger-vein images. The pixelwise curvature magnitude and principal orientation are derived to describe ridge-like veins and aggregated into blockwise histograms of oriented curvatures. This compact representation preserves spatial structure while achieving robustness and computational efficiency across varying conditions. Extensive experiments on public and self-collected datasets show that FVHOC achieves competitive recognition with low computational cost, suitable for real-time, resource-constrained applications.