I have studied two types of deformable models for computer graphics. The goal of the first effort (conducted with James V. Miller) was to develop models capable of extracting a closed geometric polygonal model from a volume dataset. We were interested in generating a model that could be used not only for visualization, but also for geometric calculations. The general approach, which we called Geometrically Deformed Models (GDMs), involved placing a closed polygonal model with spherical topology within the volume datset. We then allowed the geometric model to deform in reponse to local expansion forces and topological constraints, while interacting with the volume dataset. As the model expanded or contracted the individual polygons of the model subdivided or combined in order to maintain faces with a specified area. Once the deformation process came to an equilibrium, a closed polygonal suface is produced which approximates a specific iso-surface running through the volume dataset. (See Miller et al. 1990 & 1991). I was also involved in a similar project at Caltech that developed an interactive environment for extracting a model of the ventricles from an MRI scan of a human heart. (See Zhukov et al. 2002).
My initial research in this area explored level set methods for computer graphics, visualization, and computer animation. This was joint work with Ross Whitaker of the University of Utah, and Ken Museth of the Linköping University. Our first project developed a technique for 3D metamorphosis (morphing). Our technique guarantees that one object will smoothly transform into another object as long as the two objects initially overlap. The advantage of our technique is that user input is not required in order to produce a reasonable morphing result. Additional user input may be incorporated into the technique in order to provide user control of the morphing process. (See Breen et al. 2001, Breen & Whitaker 2001, and Whitaker & Breen 1998). Level set morphing has been used in a number of movie special effects sequences, for example in Scooby-Doo 2.
In our second joint project, we are developing a level-set framework for segmenting models from volumetric data. The framework consist of a variety of initialization methods that may be combined with level set deformation in order to extract closed, smooth structures from many types of volumetric datasets. (See Whitaker et al. 2001). For example, the framework has been used to extract the organs from a 12-day-old mouse embryo MRI scan, the structures of a developing frog embryo from time sequence MRI scans, a spiny dendrite from an electron tomogram, and the anistropic diffusion region from a diffusion tensor scan. (See Zhukov et al. 2001 & 2003). Many of the framework's capabilities have been encapsulated in Iris Explorer modules, which provide an extensible and easy-to-use interface.
In our last segmentation project we developed an approach to extracting a single high resolution level set model from several low-resolution, non-uniform volume datasets. (See Museth et al. 2002). Also see our work on level set models for geometric modeling.
I am currently working with Ken Museth (now at Dreamworks Animation) on extending our original level set morphing technique to provide more user/animator control over the morph. Coming soon to a movie theater near you!