Research On Image Super-resolution Algorithm Based On Non-convex Regularization

Image super-resolution technology,on the one hand,intuitively improves the visual effect,and on the other hand,it can also be used as a preprocessing method for other image processing technologies,such as image detection,classification,and recognition.The application of super-resolution technology in the fields of video processing,medical imaging,remote sensing technology,and security monitoring can not only save a lot of resources,but also meet the actual engineering needs.Many super-resolution algorithms have been proposed to recover high-resolution images and further improve image visualization for better image analysis.Among them,the Total Variational Regularization(TV)method has been proven to have a good effect in retaining image edge information.However,these methods based on total variational regularization do not take into account the temporal correlation between images.In addition,existing super-resolution algorithms ignore non-local information processing of images.This paper studies the problems of the above-mentioned super-resolution algorithms.The main work is as follows:(1)Aiming at the problem that the time-correlation between images is not considered in the method based on total variational regularization,a new TV regularization(TV2 ++)is designed in this paper to use the time-dimensional information of the images,so that The utilization rate of useful information in the image is further improved.In addition,the combination of global low-rank regularization and total variational regularization further enhances the restoration effect of image super-resolution.In addition,this paper uses non-convex exponential penalty function(ETP)to replace the kernel norm constraint of the matrix,and enhances the recovery ability of low-rank matrices.This paper proposes an image super-resolution algorithm based on ETP norm and TV2 ++ regularization.And use the alternating direction multiplier method(ADMM)to effectively solve the optimization problem.Experimental results show that the algorithm is superior to other algorithms.(2)Aiming at the problem of non-local information processing of the existing super-resolution algorithms,a new non-local low-rank prior(NLLRP)regularization method based on image global segmentation prior is proposed in this paper.A new method based on TV and NLLRP regularization is used to reconstruct the super-resolution reconstruction model.This method uses both local and non-local risk minimization principles.In addition,the entire optimization is performed under the framework of Alternating Direction Multiplier(ADMM).A large number of experiments show that this method is superior to several recent algorithms.