SEGMENTATION OF POLARIMETRIC SARDATA BASED ON THE FISHER DISTRIBUTIONFOR TEXTURE MODELING

ThePolarimetric Synthetic Aperture Radar (PolSAR) covariance matrix isgenerally modeled byacomplex Wishart distribution. Fortextured scenes, theproduct modelisused andthetexture component isoften modeled byaGammadistribution. Inthis paper, authors propose tousetheFisher distribution fortexture modeling. FromaFisher distributed texture component, wederive thedistribution ofthecomplex covariance matrix andwepropose toimplement theKummerU distribution inahierarchical segmentation andahierarchical clustering algorithm. Segmentation andclassification results areshownonsynthetic images andonESARL-band PolSAR data overtheOberpfaffenhofen test-site. Index Terms-Polarimetric SARimages, Fisher distribution, Texture, KummerU, Classification, Segmentation