基于非局部概率块的邻域自适应SAR图像相干班抑制算法

new neighborhood-adaptive non-local (NL) despeckling filter is proposed in this paper. An adaptive and point-wise fashion neighborhood that limits the bound of weighted pixels is designed, which is determined by an adaptive directional scales set and a new automatic similarity threshold. The set of adaptive directional scales constructs a rectangular neighborhood and the optimal scale is obtained with the proposed similarity threshold. The presented similarity is based on the probabilistic patch based similarity (PPB-similarity) measurement and deduced with a statistical Monte Carlo method. Experiment results show that our method can not only provide superior speckle removal when compared to probabilistic patch based non-local (PPB-NL) filter with fixed neighborhood, especially for its non-iterative version, but also show good performance in preserving details and texture information.