Finger-Knuckle-print recognition using dynamic thresholds completed local binary pattern descriptor

Local binary patterns (LBP) have emerged as a very powerful discriminatory texture descriptor in biometric trait analysis. Several new extensions of LBP-based texture descriptors have been proposed, focusing on improving robustness to noise by using different encoding or thresholding schemes. In this paper, a new feature set inspired by the completed local binary pattern (CLBP), known as the dynamic threshold CLBP (dTCLBP) is proposed for FKP recognition. The dTCLBP technique employs only sign and magnitude components, where the sign component is the same as the original LBP. We suggest encoding magnitude features by means of the dynamic threshold to concatenate the sign and magnitude features. The classification of this new proposed feature is performed by applying principal components analysis and linear discriminant analysis. Experiments conducted on a challenging PolyU FKP database validate its effectiveness. The results obtained indicate that the proposed technique performs well when compared to other state-of-the-art methods.