Single image super-resolution using incoherent sub-dictionaries learning

Super-resolution (SR) is recently studied to solve the problem of limited resolution in electronic imaging devices. Inspired by the success of the sparse representation in image SR, this paper presents a sparse-based SR through a new dictionary learning (DL) method. Since the DL is of great importance, we analyze the deserved properties of it in the context of three aspects. In view of the analysis, we propose a two-step procedure for DL. We first partition the training samples into different subsets, and then learn an incoherent sub-dictionary for every subset. Finally, the input patches are super-resolved using their corresponding sub-dictionaries. We further discuss the applicability of our method to consumer electronics. Experimental results show that the proposed method performs better than the existing representative methods.