Non-local methods for filtering interferometric SAR datasets

Multipass DInSAR processing is a well established technique that allows accurate 3D reconstruction and monitoring of the Earth surface. Many multipass techniques, such as the Small Baseline Subset approach, carry out a spatial multilooking of interferograms to counteract the decorrelation effects and the thermal noise and, also, to reduce computational complexity through spatial subsampling, thus enabling large scale (even full frame) analysis. Multilooking, however, degrades the spatial resolution w.r.t the original full resolution SAR product and therefore does not preserve contours of structures and in general boundaries of areas characterized by different backscattering. In this paper we propose the use of more sophisticated non-local filtering algorithms for the implementation of the interferometric multilook operation to guarantee the same noise rejection power of multilooking, while preserving high-resolution detail. Different strategies are considered and discussed for the selection of the best predictor pixels. Results on real data acquired by the ENVISAT mission demonstrate the good performance of the proposed nonlocal filtering methods.