Abstract: In this study we propose a new hybrid Island model-based approach for image segmentation structured on MAS. The general purpose of this distributed approach is to diversify a set of starting images using a GA for parallel ICM processes. This Island-MAS is composed of a set of Island-agents. Each one performs ICM starting from its own initial image. We show how cooperating Island-Agents are able to estimate the labels field and find the MAP estimate. Cooperation between Island-gents is performed by exchanging good initial images through a migration strategy. The efficiency of this Island-MAS is shown through some experimental results.