Assessment of Nonlocal Means Stochastic Distances Speckle Reduction for SAR Time Series

Implementation of complex SAR speckle filtering algorithms is usually limited to experimental settings, that make use of heavy-duty computers or processing clusters. Operational applications of SAR imagery, such as those used to map flash-flooded areas, or to flag on-going deforestation, usually are not able to take advantage of these advanced filtering techniques. Here we introduce SDNLM3D, a fast and effective 3D filtering algorithm based on the non-local paradigm, suitable to be applied on cloud environments, such as the Google Earth Engine (GEE). A systematic, real-world based benchmark of more than 700 variations of the SDNLM3D revealed that the optimized version of the SDNLM3D filter outperforms the usual filters used operationally.