Abstract
In this paper, we propose a new sampling method of Markov Random Fields(MRF). Sampling is a kind of generating process for the realization. To obtain such a realization, relaxation-type sampling methods such as Metropolis algorithm or Gibbs sampler are often used. But these sampling algorithms sometimes cause the so called phase transition. Avoiding such a phenomenon, we propose a new Metropolis-like sampling algorithm. This algorithm considers the a priori information on the relative frequencies of gray-levels in the original image, and its sampling process can be described by a finite inhomogeneous Markov chain.