A Hybrid Markov Random Field/Marked Point Process Model for Analysis of Materials Images

Both Markov random field (MRF) and marked point process (MPP) models have some limitations for image analysis. While the MRF is useful for imposing local constraints, global constraints are not easily modeled. On the contrary, it is convenient to model global constraints, such as geometric shape and object interactions, within the MPP framework, but such an object-based MPP model has limited capability for imposing local constraints such as pixel-wise interactions. In this paper, we propose a combined model that incorporates both local and global constraints within a single energy function. Optimization using our model is performed using simulation schemes, including reversible jump Markov chain Monte Carlo and multiple birth and death algorithms. We also present results using iterated conditional modes for optimization. Although our model should be useful for any application that requires both global information and precise boundary localization, we consider the analysis of microscope images of materials in this paper. We present experimental results to compare our model to the MPP model for object detection and the MRF model for segmentation.