HOS-based image super-resolution reconstruction

Abstract

In this paper a novel high-order statistics (HOS) based regularized algorithm for image super-resolution reconstruction is proposed. In this method, the image is divided into various regions according to the local forth order statistics. The segmentation label is then used to determine the weighted operator of the regularization term. In this way, different regularization terms are applied depending on local characteristics and structures of the image. The proposed image achieves anisotropic diffusion for edge pixels and isotropic diffusion for flat pixels. Experimental results demonstrate that the proposed method performs better than the conventional methods and has high PSNR and MSSIM with sharper edges. © Springer-Verlag Berlin Heidelberg 2007.