Title: Weighted two-phase supervised sparse representation based on Gaussian for face recognition
Abstract: As recently newly techniques, two-phase sparse representationalgorithms have been presented, which achieve an excellentperformance in face recognition via different phase sparserepresentation, capturing more local structural information ofsamples. However, there are some defects in these algorithms:1)The Euclidean distance metric applied in these algorithms fails tocapture nonlinear structural information, leading to that theperformance of these algorithms is sensitive to the geometricstructure of facial images. 2) To select the m nearest neighborsof the test sample is achieved directly by applying sparserepresentation in training samples, which ignores priorinformation to construct the sparse representation model. In orderto solve these problems, a Weighted Two-Phase Supervised SparseRepresentation based on Gaussian (GWTPSSR) algorithm is proposedon basic of existing two-phase sparse representation algorithm, inwhich the nonlinear local information of samples is captured byexploiting effectively the Gaussian distance metric instead of theEuclidean distance metric. Besides, GWTPSSR recreatesreconstruction set from training samples in the sparserepresentation model for each test sample, making full use ofprior information to eliminate some training samples far from thetest sample. Compared with existing two-phase sparserepresentation algorithms, experimental results on standard facedatasets show that GWTPSSR has better robustness andclassification performance.