Generalized Hard Constraints for Graph Segmentation

Graph-based methods have become well-established tools for image segmentation. Viewing the image as a weighted graph, these methods seek to extract a graph cut that best matches the image content. Many of these methods are interactive, in that they allow a human operator to guide the segmentation process by specifying a set of hard constraints that the cut must satisfy. Typically, these constraints are given in one of two forms: regional constraints (a set of vertices that must be separated by the cut) or boundary constraints (a set of edges that must be included in the cut). Here, we propose a new type of hard constraints, that includes both regional constraints and boundary constraints as special cases. We also present an efficient method for computing cuts that satisfy a set of generalized constraints, while globally minimizing a graph cut measure.