Research On The Image Restoration Based On Sparse Coding

The 21st century is an information age,as an important carrier of information exchange,images have occupied an important position in people's life and development in all walks of life.The quality of images is directly related to the exchange of people's information and the acquisition of information,and it also has some influence on people's research and judgment of things.However,the captured image is impaired due to the influence of machine or human factors in the acquisition,compression,and transmission of the image.As a result,image recovery technology continues to play an indispensable role in today's human development process,and image restoration technology has been widely used in various fieldsIn mathematics,image restoration is essentially an ill-posed inverse problem.The general way to solve the problem of image restoration is to use the prior model of natural images to regularize the space of the real solution,thus transforming the ill-conditioned image restoration problem into an adaptive problem to obtain a stable solution.At present,the perspective of image restoration by this method is divided into pixel,patch and group.However,these methods have the following two main problems First,the algorithm based on pixel points has high computational complexity;Second,the patch-based algorithm ignores the structural similarity between blocksIn order to solve the above-mentioned problems,this paper takes the sparse representation prior model of images as the starting point of research,and conducts in-depth research around the color image inpainting.The experimental results show that the algorithm have a better performance than many current state-of-the-art schemes in both peak signal-to-noise ratio and visual perceptionIn this paper,group sparse coding is taken as the breakthrough point,and the concept of group is used to link the structural similarity between patches,so that the image restoration algorithm based on adaptive non-local regularization can reduce smearing phenomenon and maintain the authenticity of image details.In addition,the sparse non-local prior model of the image is incorporated into the regular term,which can further improve the quality of the image after restoration.In the experimental part,focusing on the searching window,the influence of changing the size of the searching window on the image restoration results is analyzed,and a new index of comprehensive performance and running time is defined to determine the window size.Finally,the overall performance of the algorithm is tested and compared with mainstream algorithms to verify the superiority of the algorithm.Then lp-norm is introduced based on group sparse coding,and generalized soft threshold algorithm is used to solve lp-norm minimization problem,which makes the algorithm more stable and easier to handle.In addition,in order to improve adaptive dictionary learning,an adaptive similarity block search strategy is studied to improve the accuracy of non-local similarity block selection.In the experiment,according to different pixel loss rates,the parameters are set in detail,thus improving the accuracy of parameter setting.Experimental results show that the algorithm has better performance and algorithm stability than many current state-of-the-art schemes.