Polarimetric SAR Desjpeckling Research Based On Test Statistical

Polarimetric synthetic aperture radar (Pol-SAR) than the single polarized SARcontains more polarization information, has become the new technology of remotesensing. As of Pol-SAR image speckle noise, serious affect subsequent informationextraction, so the speckle noise suppression of a Pol-SAR data follow-up informationprocessing based on. In this paper SAR and polarimetric SAR data coherent specklenoise model and the statistical distribution of systematic research and exploration, putforward three kinds of noise suppression algorithm, the main work is as follows(1) SAR image patch similarity measure formula is derived based on test statistical,combined with NonLocal Means filtering to verify the validity of this similaritymeasure.The treatment of additive white Gauss noise PCA framework to use SARdspeckle, taking into account the SAR noise model and the statistical distribution of thedifferent, using new derivated similarty formula instead of orginal Euclidean distance tobuild a sample matric of PCA. In addition, according to SAR image specklecharacteristics derive a suitable PCA coefficient shrinkage weight formula.Aftercorrection method, the simulated SAR image and the real SAR image dspeckle hasachieved very good results;(2) Considering the characteristics of polarimetric SAR data, underpolarimetric SAR multiplicative_additive noise model framework used PCA methodto despeckle PolSAR. In order to maintain polarization information, computing similarimage patch and the weight shrinkage of PCA coefficients on Span matrix. Theexperiments show that can get good effect, the greater the advantage of the effectivenessof single-polarization SAR data processing;(3) Combined with non-local means and Lee filter, proposed a NonLocal-Leepolarimetric SAR despeckle method, which estimated each pixels under MMSEcriterion using these pixels that selected similar pixels by nonlocal method. Similaritymeasure used other papper provide which derived by test statistical. Further, in order toable to select more similari pixel,we used non-square window instead of square window.The proposed method combines the advantages of non-local means and refined Leefilter.Experiments show that this method can achieve good results.