A filtering framework for SAR data based on non-Gaussian statistics and pixel clustering

摘要

We present a filtering framework based on a new statistical model for Single-Look Complex (SLC) SAR data, the Scaled Normal-Inverse Gaussian (SNIG). The real and imaginary parts of the SLC image are modeled as mixtures of SNIGs, and the clustering of the mixture components is conducted using a Stochastic Expectation-Maximization (SEM) algorithm. Model parameters are associated to each pixel according to its class, thus producing parametric images of the entire scene. A closed-form Maximum A Posteriori (MAP) filter then delivers a de-speckled estimate of the radar texture. The method is tested on RADARSAT-2 data, HV polarization, representing images of icebergs surrounded by open water off the coast of the Hopen Island (Svalbard archipelago). Post-processing, the iceberg structure is preserved and the contrast between iceberg and water is improved (as measured by the Contrast-to-Noise Ratio), showing good potential for improving iceberg visibility in open water.