Fusing Facial Texture Features for Face Recognition

Abstract

Aiming at taking full advantage of facial information both in low-frequency and high-frequency regions and further improving face recognition rate, this paper constructs a robust nonsubsampled contourlet transform local binary patterns (NSCTLBP) feature and proposes a face recognition method fusing NSCTLBP and Gabor features. Firstly, face image is decomposed by NSCT, and the LBP values of NSCT high-frequency subbands are computed to construct NSCTLBP features. Meanwhile, convolution of 2D-Gabor wavelet with face image is performed to extract Gabor texture feature in low-frequency. Secondly, Euclidean distance and eigenvalue-weighted cosine (EWC) distance are adopted to explore the similarity measurement of NSCTLBP and Gabor features respectively. Finally, the face images are matched according to the weighted similarity of NSCTLBP feature and Gabor feature collaboratively. Experimental results on Yale and ORL databases show that the proposed method has better performances than that based on NSCT feature, NSCTLBP feature and Gabor feature separately against illumination, expression, and angle variations and glasses occlusion.