Research On Iterative Denoising Method For SAR Images Based On Speckle Statistical Characteristic Preservation

In a resolution unit of Synthetic Aperture Radar(SAR)imaging system,there are multiple scattering components,Due to the different distance of each scattering component to the resolution unit,and the phase change of the scattering component caused by the scattering and propagation process,the coherent interaction of scattered waves with different phases will show a granular noise pattern in SAR imaging,which is called "speckle".The speckle is characterized by noise similarity such as chaotic,disordered and aperiodic,which reduces the interpretability of SAR image and is usually regarded as image noise.However,from the cause of speckle,it can be seen that speckle is actually the reflection of the electromagnetic wave scattering of the ground object.Furthermore,speckle to some extent reflects part of the characteristics of the structural details of the ground object.Therefore,speckle is essentially different from other image noises and should not be regarded as a useless signal simply.The statistical theory can be used to effectively characterize the SAR image features containing inherent specks.So the statistical characteristics of speckles can be described by means of statistical modeling.Based on this idea,this paper studies the iterative denoising method for SAR images based on preserving the statistical characteristics of speckles from the point of view of statistics.This method realizes speckle suppression by making the image histogram fit the statistical distribution model of a prior hypothesis better,aiming to follow the statistical distribution regularity of SAR image in the process of noise reduction,maintain the statistical characteristics of speckles,and thus retain the ground object information that may be contained in the speckles.Different ground objects have different scattering characteristics,and their speckles follow different statistical distribution regularity.In this paper,experiments are carried out on SAR homogeneous region amplitude images and heterogeneous region intensity images,respectively.(1)Homogeneous region amplitude images.Firstly,assuming that the statistical distribution function corresponding to the homogeneous region of SAR amplitude images is square root of Gamma distribution,and the parameters of the distribution function are estimated by using the pixel amplitude-values;Secondly,according to he statistical modeling results,several gray levels with larger fitting deviation of a homogeneous region are selected by comparing the statistical distribution function and the histogram;Then,the pixels taking these gray levels as their amplitudes can be spatially located in the image domain.Among of them,some of pixels are considered as abnormal pixels with respect to ratio pixel relativity measure.The amplitude of the abnormal pixel is to be filtered by the improved Lee filter.Repeat the above procedure until the histogram of filtered image fits the distribution function well.(2)Heterogeneous region intensity images.Firstly,the statistical distribution function of a given SAR image is assumed to be a priori,namely the mixture Gamma distribution,and its distribution parameters can be estimated by the pixel values of the image.Secondly,the EM(Expectation Maximization)algorithm is used to segment the SAR image based on the mixture Gamma distribution,so as to obtain the homogeneous regions in the image.Then,for different homogeneous regions,select the gray levels with larger fitting errors,calculate the densities of the pixels whose values are the gray levels according to the segmentation results,and judge whether they are abnormal pixels.Finally,the abnormal pixels are filtered by the Frost filter.Repeat above procedure until better statistical modeling results and noise reduction results can be obtained after filtering.The experimental results show that both of SAR homogenous region images and the heterogeneous region images,on the premise of obtaining good statistical modeling results and maintaining statistical characteristics of SAR images,the proposed method can not only suppress speckle noise well,but also preserve the image edge information well.The paper has 33 diagrams,6 tables and 81 references.