Variable length dominant Gabor local binary pattern (VLD-GLBP) for face recognition

Abstract

Gabor filters are one of the most successful methods for face recognition. However they dramatically increase the data volume for face representation. To extract compact and distinctive information, we propose the Variable Length Dominant Gabor Local Binary Pattern (VLD-GLBP) for face recognition. It significantly reduces the face representation data volume whereas the performance is comparable to that of the complex state-of-the-art techniques. Specifically, local binary pattern (LBP) features are first computed from the Gabor images. Then, the most frequently occurred patterns are extracted to form VLD-GLBP. Finally the distance between VLD-GLBPs is computed to realize the face image classification. The experiment results on FERET database verify the efficiency of the proposed VLD-GLBP method.