IMAGE SEGMENTATION BY MULTIGRID MARKOV RANDOM-FIELD OPTIMIZATION AND PERCEPTUAL CONSIDERATIONS

A Markov random field (MRF) approach to image segmentation is describe d. Unlike most previous MRF techniques, which are based on pixel class ification, this approach groups pixels that are similar. This removes the need to know the number of image classes. Mean field theory and mu ltigrid processing are used in the subsequent optimization to find a g ood segmentation and to alleviate local minimum problems. Variations o f the MRF approach are investigated by incorporating features/schemes motivated by characteristics of the human vision system (HVS). Experim ental results are promising and indicate that multigrid and HVS-based features/schemes can improve segmentation results. (C) 1998 SPIE and I S&T. [S1017-9909(98)00601-1].