SAR Image Despeckling Based On Non-local Sparse Optimization

Compared with optical remote sensing,Synthetic Aperture Radar has the advantages of not being limited by time and weather,and has been widely used in many fields.However,the SAR image contains speckle,which seriously reduces the visibility of SAR image,so effective suppression of the speckle is the prerequisite for the SAR image application.With the development of the sparse representation theory,the sparse representation model has not only been successfully applied to signal processing,but also played an active role in the field of SAR image despeckling.Based on the sparse representation theory,this paper will further study the method of SAR image despeckling by combining the non-local similarity and statistical characteristics of SAR image.The main research contents are as follows:(1)In the sparse representation model based on the analysis dictionary,the calculation of image coefficients involves all dictionary bases.Compared with the synthetic dictionary model which has only a small part of bases participated in the calculation of image coefficients,the analysis dictionary model is more accurate and stable.In order to take advantage of this,this paper proposes a speckle reduction method for SAR images based on the analysis dictionary learning under tight frame constraints,and this method solves the problem of insufficient protection of image textures by the synthetic dictionary method and the problem of scale blur of theanalysis dictionary.Through experiments,it can be concluded that the analysis dictionary model has higher speckle reduction accuracy than the synthetic dictionary model,and despeckled images have richer textures in heterogeneous regions.(2)SAR image have non-local similarity and rich statistical characteristic in addition to sparsity.However,these two properties are not used in the despeckling method based on the traditional sparse representation model.In order to utilize these two properties,based on the traditional sparse representation model,this paper propose a new despeckling model for SAR images based on sparse coefficient statistical estimation combining non-local similarity and statistical characteristics.This model not only uses more sparse 3D structure group coefficients as the processing object,but also uses statistical optimal criteria in the model to solve the problem of insufficient accuracy of the coefficient estimation.Experiments verify that this method has excellent performance both in the suppression of speckle and in the protection of image texture information.(3)Compared with optical image,the statistical characteristics of SAR image in different regions are significantly different.In view of this feature of SAR images,this paper proposes a speckle reduction method for SAR images based on image block clustering,this method first clusters image blocks,and then constructs different despeckling model for different clusters according to special dictionary and sparse regular items to achieve better despeckling result,which solves the inaccuracy problem caused by adopting a unified model for all image blocks.It can be seen from the despeckling results that this method has stronger suppression capability and image detail protection capability.