Figure-Ground Discrimination by Mean Field Annealing

Laurent Herault (1) , Radu Horaud (2, 3)

1 CEA-LETI - Commissariat à l'énergie atomique et aux énergies alternatives - Laboratoire d'Electronique et de Technologie de l'Information
2 LIFIA - Laboratoire d'Informatique Fondamentale et d'Intelligence Artificielle
3 MOVI - Modeling, localization, recognition and interpretation in computer vision

We formulate the figure-ground discrimination problem as a combinatorial optimization problem. We suggest a cost function that makes explicit a definition of shape based on interactions between image edges. These interactions have some mathematical analogy with interacting spin systems -- a model that is well suited for solving combinatorial optimization problems. We devise a mean field annealing method for finding the global minimum of such a spin system and the method successfully solves for the figure-ground problem.

Domaines: Synthèse d'image et réalité virtuelle [cs.GR]

Laurent Herault, Radu Horaud. Figure-Ground Discrimination by Mean Field Annealing. 2nd European Conference on Computer Vision (ECCV '92), May 1992, Santa Margherita Ligure, Italy. pp.58--66, ⟨10.1007/3-540-55426-2_7⟩. ⟨inria-00590008⟩