Speckle Filtering For Polarimetric SAR Based On Non-local Means

Polarimetric Synthetic Aperture Radar (Polarimetric SAR) can access to a variety ofpolarization combinations of the echo, and thus can improve the capability of target detection,which is an important aspect of the microwave imaging technology development, havingextensive application in the military and civilian filed. Like other microwave imaging system,the speckle appearing in PolSAR data complicates the quality of image, affects the further dataapplication of targets Polarimetric decomposition、classifying and distinguish, so it is animportant work for despecking. This dissertation conducts research on some key theory ofpolarimetric speckle filtering technology and mainly involves the non-local means application inpolarimetric SAR despeckling. The main contributions of this dissertation include:1. The polarimetric SAR image speckle formation mechanism is analyzed, the model andstatistical properties of the speckle is described. Then summarize some polarization SAR specklereduction algorithms and analysis their points. The contents of this chapter lead to the researchwork at following chapter as a foundation.2. An optimization Lee filter based on non-local means is proposed. Through analyzing thespeckle statistics for single-look and multi-look SAR image combined with Bayesian non-localmeans filter, the conclusion that the similarity function in the multilook SAR image is accord to1-look SAR image was proved in this paper. This paper analyzes nonlocal means filter methodbased on the similarity function, presents an optimization polarimetric Lee filter based onBayesian nonlocal means. This algorithm calculates the similarity coefficient of the elements inSPAN first, then the weights are carried out to get expectation and the parameter in Lee filter,averages the covariance matrices with the weights according to the similarity between theelements, finally all terms of the covariance matrix will be filtered by Lee filter. Compared withtraditional algorithm, the experimental results based on single-look and multi-look data showthat the algorithm predigests the complexity of nonlocal means, keeps the polarimetricinformation well, and it also has the advantages of effectively reducing the speckle noise andkeeping the detail.3. A new despeckling mothed based on Beyesian nonlocal means model is proposed.Combined with the complex Wishart distribution of polarimetric covariance matrices, this paperobtains the conclusion that the similarity function of polarimetric covariance matrices based onBeyesian nonlocal means is Wishart distance, presents an optimization application of the Wishartdistance to weight computing. This new algorithm averages the covariance matrices with theweights according to the similarity between the polarimetric covariancematrices, leading toefficient reduction of the speckle in polarimetric SAR image. 4. An adaptive filtering parameter selection method based on similarity betweenpolarimetric covariance matrices is researched. Filtering parameter plays an important role innon-local means, the original non-local means is not fully effectively because of using the fixedfiltering parameter, it means that some pixels in the SAR image are filtering overly while someare not filtering well. For deal with this problem, this paper proposes an adaptive filteringparameter selection method. This method can select filtering parameter adaptively based onimage intensity statistics. Experimental results have demonstrated the robustness of this method.5. A new precluded method based on polarimetric target decomposition is proposed. Wishartdistance is derived form the statistical properties of the covariance matrices, it can only measurethe statistical similarity between pixels, Then, aiming at the defect that Wishart distance onlymeasure the similarity in statistical property, a new precluded method based on polarimetrictarget decomposition is proposed in this paper. Firstly, the polarimetric property of every pixel isobtained through Yamaguchi decomposition, then the pixel in the search window would beeliminated whose polarimetric property is not close to the center pixel. Experiment results usingreal Polarimetric SAR data show that the proposed algorithm not only despeckling effectivelybut also preserves polarimetric information well.