The interpretation of complex scenes in images requires knowledge regarding the objects in the scene and their spatial arrangement. We propose a method for simultaneously segmenting and recognizing objects in images, that is based on a structural representation of the scene and a constraint propagation method. The structural model is a graph representing the objects in the scene, their appearance and their spatial relations, represented by fuzzy models. The proposed solver is a novel global method that assigns spatial regions to the objects according to the relations in the structural model. We propose to progressively reduce the solution domain by excluding assignments that are inconsistent with a constraint network derived from the structural model. The final segmentation of each object is then performed as a minimal surface extraction. The contributions of this paper are illustrated through the example of brain structure recognition in magnetic resonance images.