Review on image restoration using group-based sparse representation

Collection of non-local patches having similar structures is termed as a group, that successively used as a basic unit of the sparse representation. This creates a brand new sparse representation modeling known as group-based sparse representation (GSR). It is able to sparsely represent the natural images within the field of group that successively force the intrinsic local sparsity and nonlocal self-similarity of images in a combined framework at the same time. For every group there is a self-adaptive dictionary learning technique is used which have low complexity. Self-adaptive dictionary learning method is an alternative to dictionary learning from the natural images. A split Bregman-based technique is developed for solving the GSR-driven ℓ0-minimization problem which makes GSR tractable and sturdy. There are three modules in our work image inpainting, deblurring, and compressive sensing (CS) recovery.