Motion artifacts free image resolution enhancement exploiting image priors

Super-resolution reconstruction algorithms have been extensively studied for the last years. However, despite the progress made in this field, many issues remain to be solved. Some of them are basically omitted and their importance is trivialized. Routinely, for instance, researchers are willing to make the relative motion model simpler than it should be considered. The commonly applied non-rigid registration method being manually defined does not capture the real motion characteristics that could occur in image sequences. This work extends Iterative Back Projection (IBP) framework in several ways. It nests image priors, deblurring and a discrete dense displacement sampling for the deformable registration of high-resolution images at its core. Applying these constraints to a global optimum of the cost function can be calculated efficiently exploiting dynamic programming. It leads to the smoothness of the deformations of the image's features. This paper proposes an improved super-resolution method while making no compromise on image quality. The author experiment results confirmed the empirical observations, in particular, that the state of the art registration algorithm and blur and noise estimate procedures, as well as image priors, lead to promising results.