Image Segmentation Based on Inhomogeneous Markov Random Field and Dirichlet Process Mixture

A novel graphical model based on inhomogeneous Markov random field and Dirichlet process mixture is proposed to address the problems encountered by hidden Markov random field. It incorporates the local and global scale information. A sampling method is also devised based on Gibbs sampler. The model is investigated in the context of image segmentation and the performance is evaluated.