A unified weight learning and low-rank regression model for robust complex error modeling

One of the most important problems in regression-based error model is modeling complex representation caused by various corruptions and environment changes images. For example, robust face recognition, images are often affected varying types levels corruptions, such as random pixel block occlusions, or disguises. However, existing works not enough to solve this problem due they cannot corrupted errors very well. In paper, we address a unified sparse weight learning low-rank approximation regression model, which enables noises contiguous occlusions be treated simultaneously. noise, define generalized correntropy (GC) function match distribution. structured disguises, propose GC based rank measure matrices. Since proposed objective non-convex, an effective iterative optimization algorithm developed achieve optimal approximation. Extensive experimental results on three public databases show that can fit distribution structure well, thus obtain better recognition accuracies comparison with methods.