Broad Area Colloquium For AI-Geometry-Graphics-Robotics-Vision
Capturing and Representing Surface Detail and Appearance
June 5, 2006, 4:15PM
Although overall geometry and texture are the fundamental ingredients
of computer graphics rendering, a great deal of richness and realism
is conveyed by subtle variations in reflectance, fine-scale detail,
and subsurface scattering. Researchers have shown that it is possible
to capture such phenomena from real-world objects, but incorporating
them as first-class citizens of the graphics pipeline requires efficient,
accurate, and meaningful representations. This talk describes recent
work aimed at distilling raw measured data into flexible and intuitive
representations, using a combination of generic data analysis techniques
and domain-specific parameterizations, constraints, and heuristics.
In the case of spatially- and directionally-varying reflectance (SVBRDFs),
measured data are approximated with intuitively editable "shade trees",
which separate different scattering phenomena into individual terms while
maintaining physical plausibility through constraints such as non-negativity,
energy conservation, and reciprocity. Similarly, heterogeneous subsurface
scattering (BSSRDF) phenomena are decomposed using a domain-specific
factored representation. Finally, the statistical nature of fine-scale
geometric detail (displacement maps) is captured with spatially-varying
analogues of texture analysis and synthesis algorithms.
About the Speaker
Szymon Rusinkiewicz is an assistant professor of Computer Science at
Princeton University, having obtained a Ph.D. from Stanford in 2001.
His work focuses on acquisition and analysis of the 3D shape and
appearance of real-world objects, including the design of scanning
devices and data structures for efficient representation. He also
investigates algorithms for processing complex geometric and reflectance
datasets, including registration, matching, completion, and sampling.
In addition to data acquisition, his research interests include
real-time rendering and perceptually-guided depiction.
Back to the Colloquium Page