New directions in texture modeling using random fields with random spatial interaction

We propose a new model for textured images of real surfaces. We establish a more general theory than the one of ordinary Conditional Markov Fields that allows the strengths of the spatial interaction to be itself a random varaable. For this class of models, we establish the power spectrum and the autocorrelation function as well defined quantities and we extract new features for texture discrimination and analysis. The new set of features that resulted from this approach was applied to real images. In contrast with the traditional Markov Fields (where samples are required to be 50x50 or larger) accurate discrimination was observed even for boxes of site 16x16.