3D Mumford-Shah Based Active Mesh
Deformable mesh methods have become an alternative of choice to classical deformable models for 3D image understanding. They allow to render the evolving surface directly during the segmentation process in a fast and efficient way, avoiding both the additional time-cost and approximation errors induced by 3D reconstruction algorithms after segmentation. Current methods utilize edge-based forces to attract the mesh surface toward the image entities. These forces are inadequate in 3D fluorescence microscopy, where edges are not well defined by gradient. In this paper, we propose a fully automated deformable 3D mesh model that deforms using the reduced Mumford-Shah functional to segment and track objects with fuzzy boundaries. Simultaneous rendering of the mesh evolution allows faster tweaking of the model parameters and offers biologists a more precise insight on the scene and hence better understanding of biological phenomena. We present evaluations on both synthetic and real 3D microscopy data.
- 6.Miller, J.: Geometrically deformed models for the extraction of closed shapes from volume data. Master’s thesis, Rensselaer Polytechnic Institute, New York (1990)Google Scholar
- 9.Szeliski, R., Terzopoulos, D.: Physically based and probabilistic models for computer vision. In: Geometric Methods in Computer Vision, vol. 1570, pp. 140–152 (1991)Google Scholar
- 19.Delingette, H.: Modelisation, Deformation et Reconnaissance d’Objets Tridimensionnels a l’Aide de Maillages Simplexes. PhD thesis, Ecole Centrale de Paris (1994)Google Scholar
- 20.Loop, C.: Smooth Subdivion Surfaces based on Triangles. PhD thesis, University of Utah (1987)Google Scholar
- 21.Ciofolo, C.: Segmentation des formes guidee par des modeles en neuro-imagerie. PhD thesis, Universite Rennes I (2005)Google Scholar