Variational Model Of SAR Image Processing Based On G~0 Distribution And Its Fast Algorithm

Synthetic aperture radar(SAR)is an important tool for observing ground objects,which has the characteristics of all-weather,all-weather,high-resolution,wide range and so on.However,the inherent speckle noise in the SAR image increases the difficulty for the subsequent SAR image processing and interpretation.Therefore,SAR image denoising and segmentation have always been the hot research topic for scholars,which have important research significance.Because of the unique advantages of variational method,more and more variational models are applied to SAR image denoising and segmentation.The existing SAR image variational denoising models usually define the regularization term as total variation(TV)term,which has good denoising effect but will cause staircase effect.Besides,the data term used in the models is not fully cosider the statistical characteristics of the speckle noise and SAR image,which leads to the limitation of the denoising effect of the model;due to the limitation of data term,the existing variational segmentation model fails in SAR image segmentation and results in unsatisfactory segmentation results.In summary,the existing SAR image denoising and segmentation models need to be improved.In view of the above problems,this paper improves the variational model on the basis of summarizing and analyzing the existing methods.The main research contents and innovations are as follows:(1)Multiplicative noise can be transformed into additive noise,thus this paper first studies additive noise removal.The total curvature(TC)model is applied to additive noise removal,in view of the complexity brought by the nonconvex,nonlinear and nonsmooth curvature terms,by introducing auxiliary variables and setting penalties,the algorithm is successfully solved based on the alternating direction multiplier method(ADMM).Compared with the classical models on the image set with additive noise,it is verified that TC regularization has the advantage of maintaining image features(i.e.edge,corner,details).(2)Based on the framework of regularization method,combining the statistical characteristics of speckle noise and SAR image,the problem of direct speckle noise removal is studied.Based on the G~0 distribution,the improved data fidelity term for SAR image denoising is given.Combining it with the hybrid regularization term which unites TV and TC regularization term,and adding the edge detection function,a variational denoising model for SAR image is established.On this basis,the energy function optimization scheme and numerical solution algorithm of the model are designed by ADMM method,and the model parameters are estimated by Mellin transformation.Finally,a great deal of experiments are carried out on synthetic and real SAR image sets.Through the corresponding qualitative and quantitative analysis,the proposed method has significantly improved the denoising performance and time efficiency.(3)Based on the variational segmentation framework and the results of SAR image denoising,a new variational segmentation model of active contour is proposed and ADMM method is used to solve the model effectively.Finally,experiments are carried out on multiple synthetic images and real SAR images to show the segmentation accuracy,convergence and efficiency of the model.