RBDL: Robust block-Structured dictionary learning for block sparse representation

Abstract

Dictionary learning methods have been extensively used in different types of image and signal processing tasks. In a number of applications, the collected data/signal may have a multi-subspace structure and be perturbed with outliers. These motivate the use of robust and block-sparse signal representations. In this paper, a new algorithm for learning a block-structured dictionary in the presence of outliers is proposed. It is based on α−divergence and has the advantage of tolerating the presence of outliers. A block coordinate descent approach is adopted to obtain simple closed-form solutions for both the sparse coding and dictionary update stages. Finally, experimental results illustrating the superiority of the proposed method over some state-of-the-art dictionary learning methods, are provided.