A Parametric Two-Dimensional Random Field with Markov Properties and Variable Structure ⋆

This paper presents a parametric, two-dimensional random field with Markov properties and variable structure that is useful for statistically modeling ensembles of images and for generating pseudo-random images with desired stochastic and visual properties. The Variable Structure Random Field (VSRF) model has six parameters which provide a high degree of control over the random field statistics and appearance. In particular, the spatial structure parameter allows for a continuum between highly structured polygon fields and unstructured Gaussian Markov fields. Experimental results show that the VSRF model can accurately characterize the autocorrelation and joint entropy of real-world images, e.g., for parameterization of image processing algorithms. This paper presents a computationally efficient method to construct pseudo-random VSRF images, which are useful as inputs for assessing the simulated performance of image acquisition, processing, and communications systems across a range of images with controlled stochastic and visual properties.