Polarimetric SAR Image Speckle Filtering Based On Scene Heterogeneity

Polarimetric Synthetic Aperture Radar system has been concerned and developed for its penetration and data acquisition capability.With the development of software and hardware platforms,the polarimetric SAR systems are more widespread,and the processing and analysis applications of polarimetric SAR images are required.Among them,the speckle phenomenon is inherent part of SAR image and hinders its application on account of the coherent interference of waves reflected from many elementary scatterers.The existence of speckle makes it difficult to use a single pixel intensity value as a measure of distributed targets' reflectivity,which causes difficulties in image interpretation and for the extraction of meaningful information.Speckle filtering for Pol SAR images has become an important preprocessing step.Considering the problems of the selection for homogeneous pixels and adaptive weighted average,this dissertation took a deep research on polarimetric SAR image filtering from the perspective of scene complexity.This dissertation systematically summarized the SAR heterogeneity features and typical polarimetric SAR filtering algorithms.On this basis,an adaptive Mean Shift filter was proposed according to the coefficient of variance(CV),and the iterative estimation was performed based on a local neighborhood.Feature-based non-local mean polarimetric SAR filter was proposed by introducing the coefficient of variance(CV)and Pauli basis(PB)to combine with the framework of the nonlocal mean filtering.Based on the analysis of the heterogeneity features,this dissertation proposed a heterogeneity measurement for Pol SAR image according to the distance measure.Moreover,the heterogeneity measurement was introduced to the Pol SAR filtering.The effectiveness of the filter algorithm and the heterogeneity measurement proposed in this dissertation were proved by utilizing the simulated and real Pol SAR data.The main contributions of this dissertation are following:1)A novel filtering method introduced the coefficient of variance and Pauli basis to measure the similarity,and the two features were combined with the framework of the nonlocal mean filtering.Various Pol SAR filtering methods have been proposed.The nonlocal mean filters estimate the coherency or covariance matrix by replacing the similarity of individual pixels with the structural similarity in the image which can contain the structure information.Therefore,this dissertation proposed that the two features were utilized to measure the similarity including the coefficient of variance and Pauli basis,which were combined with the framework of the nonlocal mean filter.Then,the similarity of the features was deduced according to the test statistic.The combination of the extracted features effectively improved the estimation accuracy of the similarity weights.In addition,the influence of the distribution model and the probability density parameter were reduced.The determined smoothing parameter could achieve better performance according to the coefficient of variance.2)To effectively test the scene heterogeneity for Pol SAR data,the distance measure was introduced by utilizing the similarity between the sample and pixels,and the Pol SAR filtering was proposed based on the heterogeneity and nonlocal means.The scene heterogeneity features,such as the CV,has been widely used in Pol SAR image processing,while,which are derived from SAR images.The heterogeneity measurement contained incomplete information of the polarimetric information of the scene due to only taking intensity into account,which damped its capability to resolve fine spatial details.This dissertation proposed a method for reducing speckle based on the heterogeneity with the framework of the nonlocal mean approaches.Firstly,the heterogeneity was measured based on the distance of K distribution to distinguish the homogeneous and heterogeneous regions.In Pol SAR images,strong backscattering from point targets was significantly different from that of distributed target,which mainly caused by a few strong elementary scatterers within a resolution cell.They did not possess the typical characteristics of speckle—not random in nature.Preservation of strong returns from strong point targets and man-made structures were desirable for image interpretation and other applications.In this dissertation,various samples were gathered according to the scene heterogeneity.Subsequently,a threshold was employed to preserve the point and lines targets.Then,a new strategy was presented to adapt to the changes in the heterogeneity map,which set the weights of the nonlocal means implemented between patches based on the heterogeneity coefficient,and the filtered image was computed.To effectively test the scene heterogeneity for Pol SAR data,in this dissertation,the distance measure was introduced by utilizing the similarity between the sample and pixels.Moreover,given the influence of the distribution and modeling texture,the K distance measure was deduced according to the Wishart distance measure.Specifically,the average of the pixels in the local window replaced the class center coherency or covariance matrix.The Wishart and K distance measure were calculated between the average matrix and the pixels.Then,the ratio of the standard deviation to the mean was established for the Wishart and K distance measure,and the two features were defined and applied to reflect the complexity of the scene.From the above analysis,the similarity weight between pixels for denosing was calculated,and the multilook Pol SAR speckle filtering was performed.3)The adaptive Mean Shift filtering was proposed algorithm based on the coefficient of variation.In order to effectively select homogenous pixels,the coefficient of variation was combined with Mean Shift algorithm.The iterative filtering was performed by applying the similarity of the heterogeneity and spatial information.Meanwhile,according to the threshold of the coefficient of variation,strong targets were filtered differently from distribute targets to improve the preservation of target features.The experimental results of single and multilook Pol SAR datasets showed that the proposed method can effectively suppress speckle and further preserve the detailed and polarimetric information.