Research On SAR Image Speckle Reduction Based On Anisotropic Diffusion

Synthetic Aperture Radar(Synthetic Aperture Radar,SAR)system is a kind of high resolution imaging radar,but speckle exist in S AR image because of the coherent processing of the scattered signals.The application of SAR image is hampered by the speckle noise,which includes target recognition technology and subsequent image processing technologies.Therefore,speckle suppression has the vital significance.The algorithms of speckle suppression mainly includes spatial-domain,transform-domain,partial differential of anisotropic diffusion despeckling.This thesis launches the research in the field of partial differential anisotropic diffusion,the main contents include the following two aspects:An algorithms of speckle suppression is presented by this thesis,which is based on improved Frost filter.The decay rate of estimated spread function is relatively slow in classic speckle reducing anisotropic diffusion(Speckle Reducing Anisotropic Diffusion,SRAD)algorithm and detail preserving anisotropic diffusion(Detail Preserving Anisotropic Diffusion,DPAD)algorithm,which leads to insensitive to the edge and detail information easily lost.In consideration of deficiency,improved Frost despeckling methods was proposed to improve the estimated spread function of DPAD,meanwhile,Self-Snake model is used to smooth the noise where it is located in the edge of the image and high coefficient of variation area.The experimental results show the despeckling algorithm can protect the information of the edge and effectively reduce speckle near edges.Three anisotropic diffusion despeckling algorithms is presented by this thesis,which is based on region subdivision.The image has blocking speckle noise in homogeneous regions,which is filted by SRAD and DPAD,and the speckle near edges can not be smoothed.Considering the deficiency,three algorithms is given.(1)The region subdivision method is completed by the ratio of edge strength map,which is based on the ratio of edge strength map(Edge Strength Map,ESM).Considering the analysis of the impact of the size of the local statistics window,a filtering process uses local small-scale statistic window(small-scale window)in the homogeneous and local large-scale statistic window(large-scale window)in the edge region.Meanwhile,the direction of the diffusion function is used to smooth the noise of the edge region.(2)The region subdivision method based on Lee filtering parameters is given.Binary image is obtained by Otsu(Otsu)dealing with Lee filtering parameters.DPAD with small-scale window is used in the homogeneous,DPAD and Self-Snake model with large-scale window are used in the edge region.Therefore,the algorithm can protect the information of the edge and effectively reduce speckle near edges.(3)The region subdivision method based on improved Frost filtering parameters is given.Binary image is obtained by Otsu(Otsu)dealing with decay factor of improved Frost filtering.DPAD with small-scale window is used in the homogeneous,DPAD with large-scale window are used in the edge region.In order to solve the problem of insufficient edge area noise smoothing,self-Snake model with adaptive parameter is applied to deal with the noise of the edge,which uses large-scale window.The experimental results show the proposed three despeckling methods can effectively reduce speckle near edges and blocking artifacts in homogeneous regions.Firstly,this thesis describes the background and imaging principle of SAR images and then proposed four algorithms anisotropic diffusion in two fields.Finally,summary and outlook are made.