Segmentation of range and intensity images using multiscale Markov random field representations

A nonlinear Markov random field (MRF) model-based segmentation method that uses multiscale MRF representations is developed. The proposed method labels the surface boundaries at a variety of spatial scales while labeling the surfaces, therefore it merges the advantages of region-based and edge-based segmentation approaches. The scheme is capable of fusing boundary information obtained at multiple scales simultaneously, resulting in a robust segmentation of range and intensity images