A New MRF Framework with Dual Adaptive Contexts for Image Segmentation

This work presents a new Markov random field (MRF) framework for image segmentation by incorporating exact contexts in the label field as well as the observed data. On the one hand, the new framework presents MRF with adaptive neighborhood (MRF-AN) system to model adaptively the contextual information of the hidden label field. On the other hand, the new framework models observations via a conditional random field (CRF), which incorporates the contextual information in observed data. The new MRF framework with the dual adaptive contextual information offers several advantages over the conventional framework. In this work, we demonstrate the advantages in an application of detail preservation in image segmentation.