A Multi-agent Approach for Range Image Segmentation

Smaine Mazouzi (1) , Zahia Guessoum (2) , Fabien Michel (3) , Mohamed Batouche

1 CRESTIC - Centre de Recherche en Sciences et Technologies de l'Information et de la Communication - EA 3804
2 SMA - Systèmes Multi-Agents
3 URCA - Université de Reims Champagne-Ardenne

In this paper an agent-based segmentation approach is presented and evaluated. The approach consists in using a high number of autonomous agents for the segmentation of a range image in its different planar regions. The moving agents perform cooperative and competitive actions on the image pixels allowing a robust extraction of regions and an accurate edge detection. An artificial potential field, created around pixels of interest, allows the agents to be gathered around edges and noise regions. The results obtained with real images are compared to those of some typical methods for range image segmentation. The comparison results show the potential of the proposed approach for scene understanding in range images regarding both segmentation efficiency, and detection accuracy.

Mots clés: Image segmentation; Multi-agent systems; Range image; Artificial potential field; Agent coordination

Domaines: Système multi-agents [cs.MA]; Traitement des images [eess.IV]

Smaine Mazouzi, Zahia Guessoum, Fabien Michel, Mohamed Batouche. A Multi-agent Approach for Range Image Segmentation. 5th International Central and Eastern European Conference on Multi-Agent Systems (CEEMAS), Sep 2007, Leipzig, Germany. pp.1-10, ⟨10.1007/978-3-540-75254-7_1⟩. ⟨lirmm-00393820⟩