Permutational model of image texture

摘要

Very often computer simulation is practically the only method of image processing studies. To organize a simulation experiment one ought to have efficient simulation algorithms of random image components. Textures form a very important class of such components. They have a specific structure which can be described by their marginal probability distributions and correlations. The coloured spectrum and non-gaussian nature make the simulation of image textures rather difficult. This paper presents a simple Gibbs random field (GRF) model and a relevant simulation algorithm. It is assumed that the modelled texture is a permutation of white noise auxiliary image elements which minimizes the appropriately defined energy function of the GRF model. The modelled texture marginal probability distribution as well as the horizontal and vertical correlation coefficients are supposed to be known. They enable to define clique potentials and an energy function for the four-neighbours neighbourhood system. Four types of causality (lexicographic, meander-like, Hilbertian and Peanonian) are considered and the simulation algorithm for four different simulation paths are proposed. Numerous simulation experiments were performed and their results analysed.