Research On SAR Image Speckle Reduction Based On Transform Domain

Synthetic Aperture Radar(SAR for short),characterized by high resolution,all-time,all-weather,and high transmittance,is of particular value in civil and defense industry.However,due to the fact that SAR is a coherent imaging system,the resulting SAR image contains lots of multiplicative speckle,which is the inherent characteristic of the SAR.The existance of speckle seriously limits the followed use of it,therefore,it is necessary to improve the imaging quality of SAR images.Now there are three commonly used algorithms.The Non-subsampled Contourlet Transform(NSCT for short)in spatial-domain is mainly discussed and two categories of new algorithms are proposed in this thesis.The first category is NSCT despeckling algorithm based on anisotropic diffusion,including two algorithms:one is speckle reduction algorithm based on NSCT and self diffusion.First,a SAR image is decomposed into multiple subbands by multi-level NSCT.Then,the each subband is filtered by self-snake diffusion,so is the reconstructed image.Compared to the traditional NSCT,which has the disadvantages of reducing only the speckle on the high-pass subbands,the improved one reduces speckle and protects the ege information by taking the filtering strategy that each subband use the same strategy but different filter parameter.The other algorithm is NSCT despeckling algorithm based on self snake diffusion and L1-L2.The algorithm is likely to the first one except that the adaptive and shrinkage soft-thresholding filter was applied to the high subbands by improved L1-L2 optimization based on the local mean.This algorithm make full use of the trait of self-snake diffusion which can effectively protect and enhance large scale edge,and the trait of improved L1-L2 optimization which can reduce speckle in mooth homogeneous regions better.The other category is NSCT despeckling algorithm based on region segmentation,including two algorithms:the first one is NSCT despeckling algorithm based on LEE and Ostu.First,LEE was applied to low-pass subband,and make the resulting as the input of Ostu,then same strategy but different filter parameter was applied to edge and homogeneous regions,for these reason,the algorithm can protect the edge information and reduce speckle better.The other is NSCT despeckling algorithm based on edge strength map(ESM).First,SAR image is divided into homogeneous regions and edge regions by shrinkage ESM.Second,self-snake diffusion was applied to the low-pass subband,and when it comes to high-pass subband,self-snake diffusion was applied to the edge regions and improved L1-L2 optimization to the homogeneous regions.ESM can converte the no direction isotropic filter to the direction of the anisotropic filter.The experimental results show the proposed diffusion despeckling algorithms in the thesis can effectively reduce speckle and protect the edge information better than many classical filtering algorithms.The thesis describes the formation mechanism of SAR and statistical models of speckle,Which is followed by analysis of several classical speckle reduction algorithms.Then two new kinds of algorithm based on NSCT are proposed and a simulation is carried on.Finally,the thesis gets some general conclusions for the whole dissertation and bring up the prospect of the work in future.