Regression Based Stereo Palm Vein Extraction and Identification System

Abstract

Palm Vein Identification(PVI) systems have been attracting interests from academia, industry, and governments for their advantages such as identification accuracy and relative low costs. However, low cost Infrared (IR) camera sensors produce noisy images which degrades the robustness of these systems. This paper proposes a new PVI system that uses a mirror based stereo camera setup to increase the PVI robustness. The two images from the stereo setup are analyzed with a new vein extraction method that uses Support Vector Regressors (SVR). The junction points of these images are compared to find junction disparities for an added 3D biometric feature. We collected a dataset of PVI images from volunteers to validate the system and we also compared parts of the proposed system on standard datasets. The overall results are promising and we will continue testing new stereo PVI image analysis methods in the future.