The Biometric Recognition System Based on Near-Infrared Finger Vein Image

Hui Ma,Na Hu,Chunxin Fang

DOI: https://doi.org/10.1016/j.infrared.2021.103734

IF: 2.997

2021-01-01

Infrared Physics & Technology

Abstract:It is a difficult task to extract vein features accurately since the finger-vein images captured by near infrared light are always poor in quality. This paper proposes a novel finger vein feature representation scheme based on pyramid histograms of oriented gradients and local phase quantization. As the vein networks consist of abundant texture and orientation features, a texture feature description operator at various scales is employed on the finger vein image to reduce the effects of geometric deformation occurred image acquisition due to the different posture and position of fingers. To solve the adverse effects of image blurring caused by uneven illumination, local phase quantization is then introduced to extract vein features. Finally, the above mentioned extracted two kinds of texture characteristics of vein image are fused at feature level by concatenated histograms to obtain accurate vein feature named pyramid local phase quantization histogram (PLPQ). In this way, we encode the vein image information not only in frequency domain but also among different orientations and scales. We perform rigorous experiments on two publicly available databases named FV-USM and MMCBUN, and the results of the experiments reveal that proposed fusion system can make promising improvement of finger vein recognition performance.