Image resolution enhancement with hierarchical hidden fields

In any image processing involving images having scale-dependent structure, a key challenge is the modeling of these multi-scale characteristics. Because single Gauss-Markov models are effective at representing only single-scale phenomena, the classic Hidden Markov Model can not perform well in the processing of more complex images, particularly near-fractal images which frequently occur in scientific imaging. Of further interest is the presence of space-variable, nonstationary behaviour. By constructing hierarchical hidden fields, which label the behaviour type, we are able to capture heterogeneous structure in a scale-dependent way. We will illustrate the approach with a method of frozen-state simulated annealing and will apply it to the resolution enhancement of porous media images.