Comparison of neural-network and model-based texture classification

Addresses the problem of the classification of clutter textures in coherent imaging, e.g. synthetic aperture radar. The major difficulty of such classification lies in defining a model by which the image may be interpreted. Progress has been made in representing such clutter textures in terms of correlated K-distributed noise. An alternative approach which is non-committal about the form of the texture is to use neural network methods which learn the underlying model from training data. The authors compare the performance of a neural network approach with a model-based one used for the classification of artificial textures generated using a correlated K-distribution noise model. This serves as a calibration of the neural network in well-characterised circumstances