This paper presented a novel robust regularized coding (RRC) model and an associated effective iteratively reweighted regularized robust coding (IR3 C) algorithm for robust face recognition (FR). One important advantage of RRC is its robustness to various types of outliers (e.g., occlusion, corruption, expression, etc.) by seeking for an approximate MAP (maximum a posterior estimation) solution of the coding problem. By assigning adaptively and iteratively the weights to the pixels according to their coding residuals, the IR3 C algorithm could robustly identify the outliers and reduce their effects on the coding process. Meanwhile, we showed that the l2-norm regularization is as powerful as l1-norm regularization in RRC but the former has much lower computational cost. The proposed RRC methods were extensively evaluated on FR with different conditions, including variations of illumination, expression, occlusion, corruption, and face validation. The experimental results clearly demonstrated that RRC outperforms significantly previous state-of-the-art methods, such as SRC, CESR and GSRC. In particular, RRC with l2-norm regularization could achieve very high recognition rate but with low computational cost, which makes it a very good candidate scheme for practical robust FR systems.
Yang, Meng, et al. “Regularized robust coding for face recognition.” IEEE Transactions on Image Processing 22.5 (2013): 1753-1766.