Matrix Regression-Based Classification for Face Recognition

Partially occlusion is a common difficulty arisen in applications of face recognition, and many algorithms based on linear representation may pay attention to such cases. In this paper, we consider the partial occlusion problem via inner-class linear regression. Specifically, we develop a matrix regression-based classification (MRC) method in which every sample from the same class are represented as matrices instead of vector and adopted to encode a probe image under. In the regression step, a L21-norm based matrix regression model is proposed, which can efficiently depress the effect of occlusion in probe image. Accordingly, an efficient algorithm is derived to optimize the proposed objective function. In addition, we argue that the corrupted pixels in probe image should not be considered in decision step. Thus, we introduce a robust threshold to dynamically eliminate the corrupted rows in probe image before making decision. Performance of MRC is evaluated on several datasets and the results are compared with those of other state-of-the-art methods.