Single Sample Face Recognition Based on DCT and Local Gabor Binary Pattern Histogram

Abstract

For years, researchers in face recognition area have been representing and recognizing faces based on subspace discriminant analysis or statistical learning. Nevertheless, for single sample face recognition these approaches are always suffering from the generalizability problem because of small samples. This paper proposes a novel non-statistics features extraction approach based on fusion of DCT and local Gabor binary pattern Histogram (LGBPH). The global and low frequency features are obtained by low frequency coefficients of discrete cosine transform (DCT). The local and high frequency features are extracted by LGBPH. To integrate the global and local features, the final recognition can be achieved by parallel integration of classification results of the global and local features. In DCT and LGBPH, training procedure is unnecessary to construct the face model, so that the generalizability problem is naturally avoided. The experimental results on ORL face databases show that the global face and local information can be integrated well after level fusion by global and local features, which improve the performance of single sample face recognition.