| Synthetic aperture radar(SAR)is widely used in military and civil fields because of its excellent performance against environmental interference.However,SAR image obtained by synthetic aperture radar with coherent imaging principle will contain a large number of speckle noise,the speckle noise of SAR image will bring great inconvenience to the SAR image research,so after getting SAR image,the first thing to do is suppressing speckle noise on it,while retaining its original basic information,including texture detail and edge information.As the nonlocal similarity of SAR images is widely researched and used,the method of SAR image despeckling has entered a new period.This thesis summarizes the despeckling method in the past,especially the method used the sparse characteristics and the nonlocal similarity of the image,and based on this,proposes the despeckling method based on structural sparse optimization of SAR image,mainly using the statistical characteristics of SAR images,the internal structure characteristics of despeckling research,the main work is as follows:(1)The traditional 3D transform domain SAR image despeckling method tends to suppress speckle noise in smooth regions ineffective,and even produces artifacts,this thesis based on the traditional 3D transform SAR image despeckling sparse constraint model,aims at the problem of the speckle noise in homogeneous areas and the artifacts,combined with the nonlocal similarity of SAR image in different areas,a new nonlocal constraints is introduced,and using weighted average value of the nonlocal similar group sparse coefficients of the target group to obtain nonlocal estimates,using patch reordering to simplify the similar images searching,and the Bayesian maximum posterior probability formula is used to obtain the relevant parameters,the final results of the despeckling obtained using split Bregman iterative algorithm and threshold shrinkage.(2)This thesis uses the low-rank characteristic of similar image patch set in the image,to avoid the nuclear norm constraint method and its threshold shrinkage method cannot constrain and estimate the singular values properly,using the nonconvex weighted norm to constrain the singular value coefficient of similar image patch set,and the estimated singular value coefficients are obtained through generalized soft threshold method,then reconstructs the image through the estimated singular value and iterates the process to get the final results.(3)To solve the general SAR image despeckling method cannot achieved good results in all areas in the image,this thesis based on the statistical characteristics of SAR images,dividing the internal SAR image into homogeneous region,heterogeneous region and mixed region through the likelihood ratio of each pixel as the center of the window,and calculating the weight for each pixel of each region according to the likelihood ratio,then weighted combining good despeckled results of the different region,and obtain the one good despeckled result of all regions. |
