A Fast Superresolution Image Reconstruction Algorithm

In a previous paper we have proposed two new superresolution image reconstruction algorithms, based on a non-parametric numerical integration Bayesian inference method, the Integrated Nested Laplace Approximation (INLA). Despite achieving superior image reconstruction results compared to other state-of-the-art methods, such algorithms manipulate huge matrices (although sparse). Therefore, the demand for memory usage and computation is high. In this paper, review such algorithms, solving these problems through relaxing one equation in the original mathematical model and involving the high-resolution (HR) image in a Torus. The result is a meaningful reduction in the computation cost of such algorithms and in the dimensions of the matrices handled as well (from n2-by-n2 to n-by-n, the size of the HR image). The result is a new algorithm, much faster than its previous version and other meaningful state-of-the-art algorithms.