Locally adaptive super-resolution through spatially variant interpolation

Systems that do not meet the requirements of the sampling theorem produce images corrupted by aliasing. Higher resolution images are attainable by unfolding aliased spatial frequencies. Multiple-image super-resolution has seen much attention in the literature though with no clear optimum algorithm for many real-world applications. We propose a method of multiframe super-resolution using a set of convolutional sinc kernels, tailored to the specific shifts between images, capable of resolving up to the diffraction limit. We demonstrate our method for the case of global shifts before we treat a pixel-level super-resolution.