Hierarchical region based Markov random field for image segmentation

In the pixel based multiscale Markov random field(MRF) model, a sequence of MRF was hierarchically defined on the multiple spatial resolutions which might suffer from the deficiency of modeling the large range of interaction. In order to overcome such a problem, we attempt to introduce the region based multiscale MRF model, in which hierarchical MRF model is defined over multiresolution image segmented regions. Based on regional multiscale MRF model, supervised image segmentation algorithm is presented and experiment on natural scene image demonstrate the better performance than the pixel based multiscale MRF model.