Data driven texture segmentation of SAR imagery

The authors compare the performance of a k-distribution model and a neural network on the desired classification for real SAR data. The same comparison is undertaken on a simulated data set to judge how closely the neural network and model based classifications approach the information limit. The aims are: to show that setting a specific segmentation goal allows segmentations to be produced which closely match those produced by eye; to compare the performance (on real and artificial data) of a model based approach to classification with that of a nonlinear adaptive filter; and to attempt to determine the measures which convey the information specific to the classification and segmentation task considered