Single-image super-resolution reconstruction based on global non-zero gradient penalty and non-local Laplacian sparse coding

• Global non-zero gradient penalty is developed to reconstruct the edge component.
• Non-local Laplacian sparse coding is developed to reconstruct the texture component.
• Global and local function is developed to further improve the imageʼs quality.

Methods based on sparse coding have been successfully used in single-image super-resolution reconstruction. However, they tend to reconstruct incorrectly the edge structure and lose the difference among the image patches to be reconstructed. To overcome these problems, we propose a new approach based on global non-zero gradient penalty and non-local Laplacian sparse coding. Firstly, we assume that the high resolution image consists of two components: the edge component and the texture component. Secondly, we develop the global non-zero gradient penalty to reconstruct correctly the edge component and the non-local Laplacian sparse coding to preserve the difference among texture component patches to be reconstructed respectively. Finally, we develop a global and local optimization on the initial image, which is composed of the reconstructed edge component and texture component, to remove possible artifacts. Experimental results demonstrate that the proposed approach can achieve more competitive single-image super-resolution quality compared with other state-of-the-art methods.