Robust single image super-resolution based on gradient enhancement

In this paper, we propose an image super-resolution approach based on gradient enhancement. Local constraints are established to achieve enhanced gradient map, while the global sparsity constraints are imposed on the gradient field to reduce noise effects in super-resolution results. We can then formulate the image reconstruction problem as optimizing an energy function composed of the proposed sharpness and sparsity regularization terms. The solution to this super-resolution image reconstruction is finally achieved using the well-known variable-splitting and penalty techniques. In comparison with the existing methods, the experimental results highlight our proposed method in computation efficiency and robustness to noisy scenes.