RL Image Restoration Based On Modulation Kernel And Extrapolation

At present,the image in the process of acquisition,transmission and storage will be affected by many factors,such as fuzzy,distortion,noise and so on.Image recovery is to process the degraded image to restore it to a clear image.In many applications,high-definition,high-quality images are needed.Therefore,it is of great significance and value to study image restoration in areas such as high-definition images.In the field of image restoration,the restoration algorithm can be divided into two categories:linear and nonlinear.Among them,the application of nonlinear restoration algorithm is more and more widely.Compared with the linear filtering method,the nonlinear iterative algorithm can achieve better recovery effect,But at the same time such algorithms also have their flaws,such as the amount of calculation,the computer hardware has a higher demand.Richardson-Lucy algorithm(RL algorithm)is one of the most widely used image restoration algorithms.The RL algorithm is researched independently by Richardson and Lucy in the field of optical image processing and astronomical image processing.At the same time Richardson-Lucy algorithm is also a nonlinear iterative algorithm.Compared with other restoration methods,such as inverse filtering method,Wiener filter method and constrained least squares method,RL algorithm has better recovery effect in noisy environment.However,the RL algorithm can not be avoided in the process of image restoration because of the increase of the number of iterations:(1)the noise expansion,reduce the signal to noise ratio,affect the image recovery effect.(2)the problem of slow convergence of iteration.Therefore,it is of great significance to improve the signal-to-noise ratio,that is,to suppress the noise and to accelerate the convergence speed of the RL algorithm.It is of great significance in the field of atomic force microscopy,astronomical images and other blurred images.RL image restoration method.In this thesis,we study the two problems in the iterative process of RL algorithm.For the problem of noise expansion,we introduce the modulation kernel method.The modulation kernel method is a kind of local gradient information which can be obtained by analyzing the local gradient information of the image,so that the image can be obtained by smoothing the local smoothing parameters.At the same time,the calculated kernel function can adjust the size and shape of the kernel function adaptively according to the local structure of the image.This adaptive characteristic can make the restoration of the image better.However,the experimental results show that the denoising efficiency of the algorithm is much lower in the face of large pixel image.Therefore,the modulation kernel is improved and the speed of the algorithm is improved.And the kernel kernel function of the modulation kernel is a partial gray-scale covariance matrix.In this thesis,the acquisition of the covariance matrix is the singular value decomposition.Therefore,the improvement of the modulation kernel is the singular value Method of improvement.In terms of modulation kernel improvement,we can make it parallel computing,thus increasing the speed of the algorithm.In order to solve the problem of slow convergence of the iteration,we study the polynomial extrapolation technique,and use the method of discontinuous extrapolation to set the RL algorithm once for each iteration to improve the RL algorithm.In this thesis,the algorithm is used to reconstruct the image by using the modulation algorithm,and then the RL algorithm is used to iterate the image,and the image is reconstructed by the intermittent extrapolation method.Finally,this thesis validates the improved restoration algorithm and compares it with the effect of restoration and iteration.The comparison result is more convincing in the form of graphs,which proves that the restoration effect of the image has been improved obviously.The proposed restoration algorithm is feasible.