ABSTRACT
Super resolution (SR) is the reconstruction of a high-resolution image with more detail from a series of low-resolution (LR) images that can either be acquired from different viewpoints or from the same view point at different times. This is possible because every LR image may contribute some information that is not in other LR images. Example applications include satellite HR image reconstruction both for panchromatic images and multispectral images, long-range camera images affected by atmospheric turbulence, multispectral classification [1], to name a few. Although many SR algorithms have been proposed, most of them suffer from several impractical assumptions: for example, that any shift or rotation between LR images is global, that the motion occurring between LR images is known exactly, or that the LR images are noise-free. However, the imaging procedure is generally much more complicated, and may include local warping, blurring, decimation, and noisecontamination. These difficulties have led us to develop a novel SR algorithm in order to avoid them and to cover all real applications.
