AN IDENTIFICATION APPROACH FOR 2-D AUTOREGRESSIVE MODELS IN DESCRIBING TEXTURES

A technique for identification of the autoregressive components is developed to describe textural regions of digital images by a general class of two-dimensional autoregressive model. Since for such modeling the feature spaces for different classes are different, a classifier and a recognition scheme using the parameters of such models are presented. Examples with synthetic textures are presented to illustrate that the model identification is appropriate. The classification scheme is illustrated with real periodic and aperiodic textures. The overall misclassification rate is about 2.4%. © 1991.