Texture characterization via sub-band predictive models

Textural information can efficiently describe the content of a remote sensing image. In this paper we present the characterization of textures via a conditional probability model for the magnitudes of the wavelet coefficients. We select the optimal model complexity using the Bayesian formalism. The order of the model and the associated model parameters can be used for texture segmentation and classification, or as an index in an image database for content-based retrieval.