Image segmentation of very large and complex microscopy images is challenging. Images are highly variable and segmentation algorithms need to be robust, fast and able to incorporate a variety of features and constraints into the segmentation model. In this study we propose a graphical model based image segmentation framework that combines the information in image regions with information about their boundaries in a probabilistic formulation. This model can be thought of as a posterior class-conditional probability density function based on above information and the image observation. We apply our approach to stained histological sections of brain or spinal cord, where axon segmentation allows quantitative measurements from closely packed axons, to elucidate the various physiological mechanisms underlying contrast in diffusion-weighted magnetic resonance (DW-MR) images.