A Novel Regularized Edge-preserving Super-resolution Algorithm

Using super-resolution (SR) technology is a good approach to obtain high-resolution infrared image. However, Image super-resolution reconstruction is essentially an ill-posed problem, it is important to design an effective regularization term (image prior). Gaussian prior is widely used in the regularization term, but the reconstructed SR image becomes over-smoothness. Here, a novel regularization term called non-local means (NLM) term is derived based on the assumption that the natural image content is likely to repeat itself within some neighborhood. In the proposed framework, the estimated high image is obtained by minimizing a cost function. The iteration method is applied to solve the optimum problem. With the progress of iteration, the regularization term is adaptively updated. The proposed algorithm has been tested in several experiments. The experimental results show that the proposed approach is robust and can reconstruct higher quality images both in quantitative term and perceptual effect.