Region-Based Super-Resolution Using Multiple Blurred and Noisy Undersampled Images

Abstract

Super-resolution is the process of combining multiple low-resolution images to produce a higher resolution image. Because the super-resolution problem is an ill-posed one, many regularized algorithms have been proposed. Those algorithms usually use the same regularization term for a whole image. However, since an image generally contains various regions of different characteristics, simple regularization is not good enough for the prospective result. In this paper, we propose a region-based super-resolution algorithm to apply a suitable regularization term for each region. In the algorithm, the image is divided into homogeneous and inhomogeneous regions. According to the type of region, we apply different filters for regularization. Regularization parameters are also adaptively determined during the iteration. Simulation results show that the proposed algorithm is superior to the conventional algorithm in terms of objective quality as well as subjective one.