Supervised radiometric and textural segmentation of SAR images

A radiometric and textural per-pixel segmentation method for single channel SAR images is proposed, which takes explicitly into account the probability density function of the imaged scene. This method makes an extensive use of adaptive preprocessing methods (gamma-gamma MAP speckle filtering, features detection by the ratio detectors, local statistics refined computation), in order to ensure good classification accuracy as well as fair preservation of the spatial resolution of its final result. Error rates prediction allows the authors to identify distinguishable classes during the training step, thus taking maximum profit of the information provided by the SAR, and saving computation time in trials.<
>