On the part-whole hierarchy in vision an example of modeling by tree-structured Markov random fields

摘要

Models based on stochastic grammars can accommodate a priori information at multiple levels of a knowledge hierarchy. Such models form the basis for the most successful existing speech recognition systems. The purpose of this paper is to argue for further study of this approach to vision problems. We use tree-structured Markov random fields to demonstrate the importance of feed-back signals through some simple segmentation experiments.