Research On Restoration Model And Algorithm Of Impulse Noise Image

Digital images are inevitably disturbed by various types of noise during acquisition and transmission,such as Gaussian noise,impulse noise,Rayleigh noise and multiplicative noise.There are a number of approaches to the impulse noise image recovery problem,including filtering methods,variational methods and deep learning methods.Among them,the variational method is a method based on a minimalist energy function that includes a data fidelity term and a total variational regularization term.In general,convex models are effective in recovering images with low levels of impulse noise,but not so effective in the case of high noise.In recent years,nonconvex models have received much attention for their ability to preserve image edges and handle high noise.Based on the existing work,this paper proposes three impulse noise image recovery models and designs corresponding algorithms to solve them.The whole paper is divided into five chapters,as follows:The first chapter introduces the research background,current research status and relevant preparatory knowledge of impulse noise image restoration and describes the main research content of this paper.The second chapter proposes a new two-stage convex optimization model for impulse noise image recovery.The alternating direction method of multipliers(ADMM)and the proximal alternating direction method of multipliers are used to solve the TV regularization model based on the preservation of details at noiseless pixel points.Numerical experiments show that the proposed method performs better in subjective visual and objective evaluation compared to other methods.The third chapter proposes a new nonsmooth nonconvex image recovery model that can effectively recover images contaminated by blur and impulse noise with flexible prior information introduction mechanisms,such as box constraints or kernel norm.For the new model proposed in this chapter,the Proximal Linearized Minimization(PLM)algorithm is used to solve it,and the ADMM algorithm is applied to solve the corresponding subproblems.In particular,the global convergence of the proposed algorithm is proved on the basis that the objective function satisfies the Kurdyka-(?)ojasiewicz(KL)property.Numerical experiments show that the proposed model outperforms the two models compared numerically and in terms of visual images.The fourth chapter proposes an efficient iterative algorithm for solving the nonconvex model used to recover the impulse noise contaminated images.In contrast to existing algorithms,this algorithm is a fully split algorithm that does not involve any subproblem.This chapter solves this problem using the Proximal Linearized Alternating Direction Method of Multipliers(PLADMM).Numerical experiments show the effectiveness of the algorithm.The fifth chapter the conclusions and outlook of the paper are given.