Directionlet Based SAR Image Despeckling

The shortcoming of the microware mechanism for SAR images, there are someperiodic and superimposed coherent signal that is speckle noise have arose up, whichhas the bad affect on the image quality and increases some difficulties for subsequentprocessing of SAR images, such as images segmentation, object identification, and soon. Therefore, removal of SAR image speckle is an indispensable step.According to the charactertistic of statistical for SAR image and the study ofmulticscale geometric analysis, especially in Directionlet transform which hasanisotropy and multi-directional characteristics. On the questions of SAR imagesdespeckling, several novel SAR image speckle reduction methods are proposed. Themain task of this paper can be summarized as the following three aspects:(1) A novel hybrid model in nonsubsampled Directionlet transform domain isderived, which is suitable for the multiplicative speckle removing. The new modelcombines hybrid model with neighborhood system. In fact, the hybrid has parametercomponents and non-parameter components. As for neighborhood system, nine differentneighborhood shapes are assumed. The new model can be effectively applied to thethreshold value estimate to the redundant nonsubsampled Directionlet transformcoefficients, which can highly be used in SAR image despeckling.(2) A new method which is based on nonsubsampled Directionlet transform andfusion rules is proposed for the SAR image despeckling. Divide SAR image tohomogeneous region and the target region. Then determine the direction of each pixel inthe target region. In the nonsubsampled Directionlet transform domain, by the means ofBayesian non-local means for transform coefficients SAR image to the specklereduction. And then, as to the coefficients need to use different fusion strategy toprccess. At last, the final speckle reduction SAR image can be obtained by makinginverse transform to the fusion coefficients.(3) A novel algorithm which is based on Dictionary learning and ProbabiliticPatch-Based (PPB) is derived, used for removing the speckling of SAR image. Theobjective function for the algorithm is based on the statistical characteristics of the SARimage. First, the reprocessed SAR image can be obtained by using PPB filter, and theresulting image is regarded as the member of the objective function. Finally, thedespecking SAR image can be acquired by using the method of the dictionary learning. The algorithms proposed in this paper are compared with classical despecklingmethods to processing the real SAR images in the experiment. The results show that thethe proposed algorithms have a better effect on speckle reduction both in visual aspectsand evaluation indicators.This thesis work is supported by the National Natural Science Foundation of ChinaYouth Project (61001206), China Postdoctoral Science Foundation funded specialprojects funded (2012T50799), China Postdoctoral Science Foundation (20110491650)and the Doctoral Program of Higher Specialized Research Fund (20100203120005).