Research on Non-Local Regularization Model and Algorithms for Image Super-Resolution Reconstruction

A non-local regularization energy functional and variational framework for image super-resolution reconstruction is proposed using the conception of non-local discontinuity of imager.The fundamental relationships between a class of generalized neighborhood filter,such as bilateral filter,and the classical variational PDE models are theoretically analyzed in this framework.The Euler-Lagrange Equation with integral formulation is derived for nonlocal variational minimizations.Some important properties of the steepest descent flow in this framework are also proved.Based on the graph theory,the authors give a novel adaptive and weighted iterative algorithms for image super-resolution.In the end of paper,different examples of non-local regularization energy functional is used to image de-noising,image de-masaic,and image super-resolution.Experiments show that the non-local regularization functional has better performance than the classical regularization functional under the same potential function,and the Peak SNR values are averagely increased by about from 0.5 to 1.0 dB.