Digital Deformable Model Simulating Active Contours

François de Vieilleville (1) , Jacques-Olivier Lachaud (1)

1 LAMA - Laboratoire de Mathématiques

Deformable models are continuous energy-minimizing techniques that have been successfully applied to image segmentation and tracking since twenty years. This paper defines a novel purely digital deformable model (DDM), whose internal energy is based on the minimum length polygon (MLP). We prove that our combinatorial regularization term has convex-like properties: any local descent on the energy leads to a global optimum. Similarly to the continuous case where the optimum is a straight segment, our DDM stops on a digital straight segment. The DDM shares also the same behaviour as its continuous counterpart on images.

Domaines: Vision par ordinateur et reconnaissance de formes [cs.CV]; Traitement des images [eess.IV]

François de Vieilleville, Jacques-Olivier Lachaud. Digital Deformable Model Simulating Active Contours. International Conference on Discrete Geometry for Computer Imagery, Sep 2009, Montréal, Québec, Canada. Accepted. To appear. ⟨hal-00413684⟩