Structured regularized robust coding for face recognition

The sparse representation based classifier (SRC) has been successfully applied to robust face recognition (FR) with various variations. To achieve much stronger robustness to facial occlusion, regularized robust coding (RRC) was proposed by designing a new robust representation residual term. Although RRC has achieved the leading performance, it ignores the structured information (i.e., spatial consistence) embedded in the occluded pixels. In this paper, we proposed a novel structured regularized robust coding (SRRC) framework, in which a weight value is assigned to each pixel to measure its importance in the coding procedure and the spatial consistence of occluded pixels is exploited by the pixel weight learning (PWL) model. Efficient algorithms were also proposed to fast learn each pixel's weight value. The experiments on face recognition in several representative datasets clearly show the advantage of the proposed SRRC in accuracy and efficiency.