Bayesian Texture Based Analysis of Hr Slc Sar Images

The Bayesian approach is a promising method for modelbased signal analysis. It was previously used on detected radar images for model based despeckling and feature extraction. We propose an extension on Single Look Complex (SLC) High Resolution (HR) Synthetic Aperture Radar (SAR) images. The information contained in the phase is reflected in the second order statistics and it is important for texture characterization. The SLC data, generally modeled as circular complex Gaussian, is assumed to be modeled by a complex Gauss-Markov Random Fields (GMRF). An efficient parameter extraction for texture characterization is important in order to create an alphabet of plausible primitive feature for image labeling. The affectation of the phase correlation on parameter estimation is explored. The results are demonstrated on E-SAR SLC HR images.