Abstract
Super-resolution can be applied to different fields in many aspects. Generally, super-resolution is used to solve the blurring effect resulted by interpolation, however, majority of the algorithm requires high computation. We restored the distort signal based on the image formula process model. Therefore, we proposed a non-iterative deconvolution with gradient priors as a restraint condition to reconstruct a high resolution image. Hereby, we shall propose two ways of getting the gradient priors which are the filter-based and the learning-based respectively. To require the gradient priors using the learning-based can get a better reconstruct quality. In addition, this algorithm can also apply directly to the videos without considering the relationship of the temporal coherence.