pairwise constraints

Image segmentation remains a challenging problem in computer vision.  Among the various techniques, graph-based methods have become increasingly popular, with modern approaches often incorporating some form of user interaction.  Following this trend, we propose a new approach to interactive image segmentation based on dominant sets, which generalize the concept of maximal cliques to edge-weighted graphs.  In particular, we extend this framework to handle pairwise constraints.  By expressing the user-provided scribbles as must-link and cannot-link constraints, we are abl