Multi-look Polarimetric SAR Speckle Filtering Based on Unsupervised Classification
基于无监督分类的多视极化SAR相干斑滤波

The presence of speckle is a major cause of degradation in SAR images. The Multilook Polarimetric Whitening Filter (MPWF) is an effective method on speckle reduction in multilook polarimetric Synthetic Aperture Radar (SAR) images. However, the capability of the filter is directly decided with the precision of the filter-parameter estimation. Hence, a novel adaptive windowing algorithm based on unsupervised classification is proposed here, where the classified image is chosen as the processed object and the central pixel in moving rectangular window is as the reference. Then through automatic search, the pixels around the central pixel which are in the same class are selected and used for parameter estimation. Compared to some other typical methods, this approach is demonstrated the effectiveness both on speckle reduction and preservation of texture information from the experimental results.