Applying integrated nested laplace approximation to the superresolution problem

Superresolution is a term used to describe the generation of a high-resolution image from a sequence of similar low-resolution images. In 2011 we derived a closed form to resolve the superresolution problem, thus proposing a new algorithm to generate the high-resolution image. However, the choice of an hyperparameter (λ), involved in the fusion of the low-resolution images, is still heuristically defined. Thus, to get a good value for such hyperparameter is somewhat troublesome, demanding much experience or a lot of attempts. In this context, this paper proposes a fully automatic method for choosing such hyperparameter, thus providing a fully analytical solution for the superresolution problem. In the solution it is used, by the first time in the image processing field, a new Bayesian inference method known as Integrated Nested Laplace Approximation (INLA). Several simulations, from which two results are here presented, show that the proposed algorithm performs better than other superresolution algorithms yet available in the literature.