Adaptive Non-Local SAR/PolSAR Despeckling Algorithm Based On Concern Graph

Synthetic aperture radar(SAR) is an advanced microwave technology, it has been widely applied in both civil and military fields.The using of a large number of earth observation data obtained by this technique provided an important basis for people to understand the Earth system. Because of coherent imaging mechanism of SAR image, speckle has been produced, which destroyed the target information and increased the difficulty for people to identify and determine the object. Useful information content of SAR images and polarimetric SAR data we obtained become low significantly due to the presence of noise.How to extract useful information from these data is the key to the interpretation of SAR and polarization SAR. Currently, many researchers have made a number of efficient algorithms, such as PPB, Lee, Pretest algorithms, these methods have their own advantages and disadvantages. Based on the SAR imaging mechanism, reasons for the formation of noise and full analysis for the noise model,we proposed two speckle suppression algorithms to deal with the speckle in SAR images and polarimetric SAR data. The article mainly includes two works as follow:1? We propose a new bilateral filtering algorithm combined with the degree of local homogeneity based on patches' similarity for SAR despeckling. First, we design a new homogeneity significant calculation method combined the traditional significant algorithm with the multiplicative noise model of SAR images. The using of weighted estimators try to eliminate the effect of noise on a significant and obtain more accurate homogeneous saliency map; Secondly, we develop adaptive weighting formula based on the relationship between "bilateral" and saliency map; Finally, a clean image is produced by improved bilateral filtering algorithm. The proposed algorithm has made full use of surface texture features with adaptive parameters based on the saliency of homogeneity for despeckling.The final result shows that it has a well balance between smooth the area and preserve points, lines, edges and other details.Simulation SAR image and real SAR image filtering results show the effectiveness of this method, not only the noise can be suppressed but also the image texture, structure information can be protected.2?We built a decomposition similarity model and propose a non-local algorithm based on decomposition similarity and wishart distance. In order to exploit fully the inherent polarization SAR data information and obtain its scattering characteristic information by exploding its covariance matrix and statistical information with the help of wishart distance measure similarity, a combination of both can obtain a more accurate measure of similarity than before. Then an improved polarimetric SAR non-local means filtering algorithm which use homogeneous significant to adjust adaptively the size of search window has been proposed. The experiments on three polarimetric SAR images show that the suppression of speckle in homogeneous areas and preservation in structure areas both are effective mean while the algorithm improve the lack of Pretest filter, that is, Pretest filter cannot keep the structure details well.