Sparse representation based undersampled face recognition with shared prototype–auxiliary dictionaries

Abstract

The sparse representation with auxiliary dictionary based face recognition methods have achieved significant performance in recent years. The prevailing auxiliary dictionary based methods use training dictionary and auxiliary dictionary to separate facial samples’ prototype dictionary and the intra-class variation component respectively. While in undersampled cases, training dictionary usually contains large intra-class variations, the prototype component cannot be fully separated. For this limitation, a sparse representation based classification with shared prototype–auxiliary dictionaries (SRSPA) method is proposed. In SRSPA, a shared prototype dictionary is exploited to specifically separating the prototype component. In addition, a novel dictionary learning method is proposed, which fully considers the separation ability of prototype dictionary and auxiliary dictionary. Experiments on various data sets verify efficacy of the proposed SRSPA especially in undersampled cases.