Bayesian Method for Segmentation of Sar Images in Rough Terrain

Radiometric correction is the essential prerequisite to obtain precise and valuable segmentation of remote sensing images, especially when dealing with mountainous regions where the terrain is more likely to be rough. Important applications such as snow cover segmentation have usually to be performed on images of very rough mountainous terrain where this preprocessing step turns out to be especially demanding. The knowledge of the topography of the imaged area through a digital elevation model (DEM) and of the backscatter function for the different terrain cover types are the basis for radiometric correction. Considering SAR images, the huge amount of processing for geographic and geometric calibration and registration that is needed prior to analysis is well established. Nonetheless, even assuming that these calibration and registration steps can been carried out with high precision algorithms, they are still prone to inaccuracies due to the quality of the terrain geometrical description. In the following is presented a model-based method that, exploiting the information contained both in the DEM and in the image, provides improved estimates, in a Bayesian framework, of the terrain itself and of the radiometric characteristics of the land cover.