Unsupervised image segmentation using gibbs sampler within a multiresolution framework

This work approaches the texture segmentation problem using Gibbs sampler (i.e., the combination of Markov random fields and simulated annealing) within a multiple resolutions framework with "high class resolution and low boundary resolution" at high levels and "low class resolution and high boundary resolution" at lower ones. As the algorithm descends the multiresolution structure, the coarse segmentation results are propagated down to the next lower level so as to reduce the inherent class-boundary uncertainty and to improve the segmentation accuracy. The under-segmentation problem due to the excessive inter-scale interaction in our previous work is addressed and a new neighborhood system and paradigm for inter-scale interaction is proposed to attack the problem.