Sparse media image restoration based on collaborative low rank representation

Visual quality for image is a important problem that limits the performance in media big data analysis. To address this problem, we proposed a novel media image restoration method based on sparse representation to produce better media data to be analyzed in subsequent tasks. Sparse representation is one of the significant technique for signal processing, which aims to present the signal as the combination of several special atoms chosen from a over-completed dictionary and shows some promising results in image restoration. However, without utilizing the correlation among patches in image, the conventional sparse-based methods process the patches individually, which may not be effective enough to obtain the satisfied recovery results. Hence, in this paper, to improve the performance, an extra coding constraint based on low rank representation is introduced. The latent structure among nonlocal similar patches are firstly explored by low rank representation. And then, we exploit the structure to form a extra regularization to constrain the coding approximation between each pair of patches, which is believed to be helpful to preserve the details in image. Our extensive experiments on various benchmark images validate the effectiveness and efficient of the proposed method.