Research On Structure Preserving Speckle Reduction For PolSAR Image Based On Nonlocal Means Filtering

Polarimetric synthetic aperture radar(PolSAR)system has unique advantages in observing the Earth's surface whatever time and weather.Compared with SAR,PolSAR is able to provide more target scattering information.It has extensive application prospect not only in the military field but also in the fields such as resources exploration,environmental monitoring.However,restricted to the special imaging mechanism of SAR,PolSAR images are corrupted by speckle noise.The presence of speckle increases the difficulty of image understanding and decreases the accuracy of subsequent image segmentation and classification.Thus,research on the despeckling algorithms of PolSAR image has great theoretical significance and high practical value.Concerning the preservation of structure information as the difficult problem in speckle suppression,this paper analyzes methods to select the homogeneous pixels before filtering.Under the framework of nonlocal means filtering,two structure preserving despeckling algorithms for PolSAR images are proposed.In conclusion,the main innovations of this thesis are as follows:(1)A new nonlocal means filter based on 3-D patch matching wavelet transform is proposed.Combining 3-D patch matching based undecimated wavelet transform with the structural similar pixels selection,the proposed filter increases the accuracy of structural similar pixels selection,and enhances the expression ability of image structure feature.Structural preserving weight function is constructed to amplify the weight of structural information in the similarity measurement between patchs.Sigma filter is used to extract the polarimetric scattering feature,which is incorporated into the despeckling process.The proposed filter can reduce the speckle effectively while retain structural and polarimetric scattering characteristics in PolSAR images.(2)A nonlocal means filter based on dual-feature clustering is proposed.The homogeneous pixel clusters are composed of pixels with similar structural and polarimetric scattering characteristics.The pixels with similar structural characteristics are selected by the patch-statistical distance based Mean-Shift clustering method meanwhile,the pixels with similar polarimetric scattering characteristics are selected by the Freeman-Durden scattering model based Wishart unsupervised classifier.Through the above two adaptive clustering methods,the selection accuracy of homogeneous pixels in PolSAR images is greatly enhanced,which impoves the despeckling property.To verify the effectiveness of the proposed despeckling algorithms in PolSAR images,simulated data and real PolSAR data are used for validation.The experiments demonstrate that:compared with other polarimetric filtering algorithms,the proposed algorithms can better preserve the structural and polarimetric information while reduce the speckle effectively.