Texture Synthesis and Unsupervised Recognition with Nonparametric Multiscale Markov Random Field Models

In this paper we present noncausal, nonparametric, multiscale, Markov Random Field (MRF) models for synthesising and recognising texture. The models have the ability to capture the characteristics of a wide variety of textures, varying from the structured to the stochastic. For texture synthesis, we use our own novel multiscale approach, incorporating local annealing, allowing us to use large neighbourhood systems to model some complex textures. The new multiscale texture synthesis algorithm also produces synthetic textures with few phase discontinuities. The power of our modelling technique is evident in that only a small training image is required to synthesis representative examples of the training texture. We also show how the high dimensional model of the texture may be modelled with lower dimensional statistics without over compromising the integrity of the representation. We then show how these models can be used for the unsupervised texture segmentation and recognition of images containing previously unseen textures; a technique useful in the practical application of recognising different terrain types from Synthetic Aperture Radar (SAR) images.