Gabor vs. GMRF features for SAR imagery classification

A comparison of the ability to discriminate among distinct regions in synthetic aperture radar (SAR) imagery using textural features based on two different methods is presented. Features are generated from Gauss Markov random field (GMRF) model parameters and from Gabor convolution energies. The discrimination ability is evaluated in terms of misclassification errors resulting from tests performed on a patchwork of different MSTAR clutter regions.