An information-theoretic characterization of Markov random fields and its applications
Raymond W. Yeung, T.T. Lee, Zhongxing Ye · 2002
We obtain a non-redundant characterization of a Markov random field (MRF) in terms of the unconditional joint entropies of the random variables involved. A canonical form for information expressions conditioning on a MRF is identified, and the dimension of the Shannon's information measures of a MRF is determined.