Research on finger vein recognition based on NSST

Abstract

In the finger vein image collection procedure, there are always two kinds of negative factors: the first is unstable illumination, the second is information loss caused by bad collection operation. There are not yet special methods for the above-mentioned question for now. We adopt the Non-Subsampled Shearlet Transform (NSST) coefficients for feature extraction, since the NSST transform domain coefficients are affected less by unstable illumination. For the question of information loss, we first introduce an improved ROI extraction method for database extension. We further propose an improved robust regression classification method for vein recognition. The experimental results show that: compared with traditional methods, our proposed method based on NSST does better in recognizing finger vein images which lack some information and are influenced by the unstable illumination.