Generalized circular autoregressive models for isotropic and anisotropic Gaussian textures.

Abstract

A new class of random field models, called generalized circular autoregressive (GCAR) models, is introduced. GCAR models have noncausal neighbors that have the same autoregressive parameter values if they are on the same circle or ellipse and that have circular or elliptical correlation structures. This model is better for modeling isotropic or anisotropic natural textures than earlier approaches to modeling of isotropic textures and can represent complex textures with a small number of parameters. Parameter estimation is also considered, and a multistep estimation algorithm is presented. Properties of estimators of GCAR models are also investigated. The efficacy of GCAR models in modeling real textures is demonstrated by synthesizing images resembling real textures by use of parameters estimated from textures selected from the Brodatz texture album. Limitations of GCAR models are also discussed.