Towards Perceptual Control of Markov Random Field Textures

Abstract

In this paper we propose a method of establishing a perceptually meaningful Euclidean space for textures generated as samples of first-order Markov Random Fields (MRF). We show how, under a definition of texture difference, within a specified neighbourhood of zero the first-order MRF parameters can be considered as orthogonal and approximately define textures in a uniform manner. We suggest that the Euclidean space established with this texture distance metric may be considered a perceptual texture space that can be used for visualisation purposes. The definition of texture difference is based on texture classification results that are consistent with human visual discrimination when experimented with a set of test images. Demonstrations of using our texture space as a preliminary visualisation tool are presented.