Unsupervised Segmentation Of Multiscale Stochastic Model And The Sar Image

The multiscale mixed distribution models(MMDM) and the multiscale autoregressive(MAR) models are investigated in this thesis, and they are applied to the unsupervised segmentation of the synthetic aperture radar(SAR) image by joining them together-the multiscale mixed distribution models as the feature extractor and the multiscale autoregressive models as the classifier. The main innovations in this thesis are: (1) given the notions of MMDM, and argumentations of the approximation ability of the MMDM, (2) samples competition algorithm and its application to the estimation of the MMDM's parameters, (3) unsupervised segmentation of SAR image by the MMDM, (4) segmentation criteria, (5) unsupervised SAR image segmentation by combining the MMDM and the MAR models, (6) the MAR's order problems.