NEW 3D GRAPHICS TECHNIQUE PROPOSED
Prof. Desbrun's new technique partitions a shape into geometric "proxies" (colored areas of the shape on the left) that can be represented as ellipses (center). They create a "blueprint" for a concise, polygonal geometric description (right).
Small triangles are gathered into optimal regions, then combined into larger ones, and finally converted into polygons. The entirely automatic process suppresses superfluous faces and results in a much tighter representation, with guaranteed approximation quality.
Prof. Mathieu Desbrun, an IMSC key investigator, has developed a new technique to generate smaller and more efficient 3D graphics files. He will present the findings at the 2004 SIGGRAPH conference, August 8-12, in Los Angeles.
The technique, called "variational shape approximation," produces highly accurate 3D graphics files many times smaller than today's methods. Presently, triangles are used to represent a 3D object. Yet, surprisingly, rare are the techniques for automatically arranging these triangles spatially to optimize the approximation quality of the surface for a given number of triangles.
As 3D scanners become commonplace, such an optimal property is crucial to avoid redundant and superfluous storage of large geometric datasets. Prof. Desbrun's technique offers a simple machine-learning based solution to this problem. The technique uses polygons as well as triangles, providing greatly enhanced flexibility and efficiency.
His collaborator, Pierre Alliez, who worked with him in 2001 as a post-doctoral fellow at USC, is now a researcher at the National Institute for Research in Information and Automation (INRIA) of France. The first author of the SIGGRAPH paper, David Cohen-Steiner, is currently a post-doc at Duke University. For the paper, go to http://www-grail.usc.edu/pubs/CAD04.pdf.