MRF and multiagent system based approach for image segmentation

Simulated Annealing (SA) and Iterated Conditional Modes (ICM) are two of the Markov Random Fields (MRF) model based appi'oaches for image segmentation. In practice, the ICM provides reasonable segmentations compared to the SA and was the most robust in most cases. However, the ICM strongly depends on the initialization phase. In this work, we develop a new approach for image segmentation based on Multiagent System (MAS) in order to produce good segmentations. We consider a set of segmentation agents and a coordinator agent. Each segmentation agent is able to segment the image by ICM starting from its own initialization. However, the coordinator agent diversifies the initial configurations using crossover and mutation operators known in the Genetic Algorithms (GAs). We can consider this model as a hybridization of ICM and GAs. The role of this hybridization is to help in the task of segmentation intensification in order to accede to good configurations.