BISPECTRAL ANALYSIS AND MODEL VALIDATION OF TEXTURE IMAGES

Statistical approaches to texture analysis and synthesis have largely relied upon random models that characterize the 2-D process in terms o f its first- and second-order statistics, and therefore cannot complet ely capture phase properties of random fields that are non-Gaussian an d/or asymmetric. in this paper, higher than second-order statistics ar e used to derive and implement 2-D Gaussianity, linearity, and spatial reversibility tests that validate the respective modeling assumptions , The nonredundant region of the 2-D bispectrum is correctly defined a nd proven, A consistent parameter estimator for nonminimum phase, asym metric noncausal, 2-D ARMA models is derived by minimizing a quadratic error polyspectrum matching criterion, Simulations on synthetic data are performed and the results of the bispectral analysis on real textu res are reported.