An Adaptive Nonlocal Mean Filter for PolSAR Data with Shape-Adaptive Patches Matching

The traditional nonlocal filters for polarimetric synthetic aperture radar (PolSAR) images are based on square patches matching to obtain homogeneous pixels in a large search window. However, it is still difficult for the regular patches to work well in the complex textured areas, even when the patch size has a small enough setting (e.g., 3 × 3 windows). Therefore, this paper proposes an adaptive nonlocal mean filter with shape-adaptive patches matching (ANLM) for PolSAR images. Mainly, the shape-adaptive (SA) matching patches are constructed by combining the polarimetric likelihood ratio test for coherency matrices (PolLRT-CM) and the region growing (RG), which is called PolLRT-CMRG. It is used to distinguish the homogeneous and heterogeneous pixels in textured areas effectively. Then, to enhance the filtering effect, it is necessary to take the adaptive threshold selection of similarity test (Simi-Test) into consideration. The simulated, low spatial resolution SAR580-Convair and high spatial resolution ESAR PolSAR image datasets are selected for experiments. We make a detailed quantitative and qualitative analysis for the filtered results. The experimental results have demonstrated that the proposed ANLM filter has better performance in speckle suppression and detail preservation than that of the traditional local and nonlocal filters.