Super-Resolution Reconstruction Using Probability Model Combined with Nonlocal Low-Rank Prior

Nowadays imaging technologies have become more and more popular.However,due to the mutual constraints from imaging equipment,environment and other factors,such as external noise in the process of acquiring images,the image resolution is generally low in the practical applications,which causes many problems.In this paper,we propose an effective image super-resolution reconstruction model based on maximum a posterior probability(MAP)and nonlocal low-rank prior(NLP).Firstly,by inputting the continuous image sequence,the similarity inside the single image and among the image sequence is used as prior knowledge,in order to improve the matching degree of similar image patches and eliminate the loss of image details.Then,the reconstruction is modeled with MAP framework,where the parameters are fitted by Gaussian distribution and Gibbs distribution,respectively,for increasing the generalization capability.Furthermore,this model estimates the singular values of the desired patches by singular values of similar patches,and suppresses the noise by low-rank truncation.Finally,to exploit the nonlocal self-similarity and low-rank nature of images,NLP regularization is adopted to regularize the reconstruction process,which introduces the local and global image information to improve the reconstruction effect.The experimental results on the standard optical flow datasets and the datasets provided by New York University and Shandong Provincial Qianfoshan Hospital show that,this proposed model based on MAP and NLP is comparable to the traditional interpolation algorithms and the excellent reconstruction-based algorithms.This method increases the average peak signal-to-noise ratio by 6.3 dB in five simulation experiments and achieves better reconstruction performance in maintaining image texture features and restoring image details.