A new approach for extracting shape from texture

Abstract

A novel way of modeling images that result from the projective distortions of homogeneous textures laid on illuminated 3D surfaces, as they are seen by a camera is presented. A Gaussian Markov random field (GMRF) is used for modeling the homogeneous planar parent texture. The projective distortions of the parent texture is a 2D Gaussian random field described by a probability distribution which is an explicit function of the parameters of the GMRF homogeneous texture model, the surface shape, and the camera model (orthographic or pinhole). The basic modeling concepts are used in extracting shape information from texture. Shape parameter estimation is posed as a maximum-likelihood estimation (MLE) problem.<>