Broad Area Colloquium For AI-Geometry-Graphics-Robotics-Vision
Geometry Images: Sampling Surfaces on Regular Grids
Monday, May 24, 2004, 4:15PM
Surfaces in graphics are commonly represented using irregular meshes, since
these approximate many shapes using fewer vertices. However, their flexible
connectivity comes at a price: most mesh operations require random memory
accesses through vertex indices and texture coordinates; filter kernels must
handle arbitrary neighborhoods; and, techniques like morphing, level-of-detail
control, and compression are complicated.
In contrast, media like audio and images are represented using regular
samplings - 1D and 2D grids. Such grids allow efficient traversal, random
access, convolution, composition, down-sampling, compression, and
synthesis. Many surface signals have now migrated into texture images. As the
cost of 3D transformations becomes negligible, one should re-evaluate whether
geometry itself would not be better represented using ordinary grids.
This talk will present several recent projects related to geometry images.
These are constructed by parametrizing the surface over a planar domain, and
resampling the surface geometry on a regular domain grid. One exciting
application area is to then exploit the highly parallel GPU rasterizer to
directly process geometry.
About the Speaker
Hugues Hoppe is a senior researcher in the Computer Graphics Group at Microsoft
Research. His primary interests lie in the acquisition, representation, and
rendering of geometric models. For his PhD work on surface reconstruction from
3D scans, he was selected as a finalist in the 1995 Discover Awards for
Technological Innovation. He subsequently developed multiresolution
representations for geometry, including multiresolution analysis [Eck et al
1995] for semi-regular meshing, progressive meshes [Hoppe 1996] for irregular
meshing, and geometry images [Gu et al 2002] for completely regular meshing.
Recent work has focused on surface parametrization, where contributions include
lapped textures, normal-shooting parametrization, geometric-stretch metrics,
hierarchical solvers, signal-specialized parametrization, and spherical
parametrization. His most recent interest is the regular sampling of surfaces
using geometry images, anticipating the unification of vertex and image
buffers. His publications include 20 papers at ACM SIGGRAPH, and he is
associate editor for ACM Transactions on Graphics. He is also the 2004 winner
of the the ACM SIGGRAPH Achievement Award. He received a BS summa cum laude in
electrical engineering in 1989 and a PhD in computer science in 1994 from the
University of Washington.
Back to the Colloquium Page