Extracting occlusion from images based on texture-synthesis methods

This paper proposes a method of region segmentation by which occlusions are extracted from a non-repeating, non-parametric texture. Based on a method of texture synthesis proposed by Efros and Leung (1999), this method of segmentation assumes the texture can be described by a Markov random field with probabilities conditioned on neighboring pixels. Instead of generating a new texture, the Markov model is used to determine the acceptance of pixels as part of the texture. This method has the advantage of not requiring any special filter preprocessing or texton processing.