Segmentation of surfaces and volumes is considered. Specifically, variational approaches are applied to solving the following two problems: (1) surface segmentation by quadric surface fitting; 2) segmentation of a 3D volume into components that can tightly be bounded by ellipsoids. A new variational formulation based on anisotropic metrics, effective computation schemes and convergence analysis will be presented.
About the Speaker:Dr. Wenping Wang is Associate Professor of Computer Science at University of Hong Kong. His research covers computer graphics, geometric computing and visualization, and he has published widely in these fields. He got his Ph.D. in Computer Science from University of Alberta in 1992. He is associate editor of Computer Aided Geometric Design, and has been program co-chair of several international conferences, including Geometric Modeling and Processing, ACM Symposium on Physical and Solid Modeling, and Pacific Graphics.
Almost all current graphics libraries and, consequently, almost all application packages to visualize graphical 3D information are based on the approach that the objects' surfaces are approximated by a set of triangles, resembling the individual shape in the best possible way. Of course the term 'best possible' is a function of the individual task to be accomplished and the triangle budget provided by the hardware or the network bandwidth in use.
The major motivation for this approach is the fact that graphics hardware over the years has been optimized to display large numbers of triangles at very high speed. The disadvantage of this development, however, is that the complete semantic representation of geometric objects is given up much too early - sometimes even at the stage of the modeling process - in favor of a triangle-based representation. As a consequence, a significant R&D effort had to be undertaken (and still is) to compact the sheer mass of triangles into units which fit into current bandwidth constraints with regard to data exchange, processing and visualization.
The talk will present - in the context of recent developments for future Digital Libraries - the Generative Modeling approach as a promising alternative to triangle-based modeling and rendering by consequently preserving and exploiting the semantic object representation from the modeling stage to the last stages of the rendering pipeline. And as recent results show generative modeling might also be very well suited to provide the basis for a new generation of metadata 'vocabularies' necessary to effectively and automatically tag and index 3D models.
It will be both a presentation and a discussion about what can be done, how it can be done, and the limitations: