An image labeling algorithm based on cooperative game theory

Many image analysis and computer vision problems can be formulated as a scene labeling problem in which each site is to be assigned a label from a discrete or continuous label set with contextual information. In this paper we present a new labeling algorithm based on game theory. More precisely, we use Markov random fields to model images, and we design an n-person cooperative game which yields a deterministic optimization algorithm. Experimental results show that the algorithm is efficient and effective, exhibiting very fast convergence, and producing better results than the recently proposed non-cooperative game approach. We also compare this algorithm with other labeling algorithms on real-world and synthetic images.