Perception-Guided Global Illumination Solution for Animation Rendering

Karol Myszkowski, Takehiro Tawara, Hiroyuki Akamine and Hans-Peter Seidel

Max-Planck-Institute for Computer Science

Appeared in the SIGGRAPH 2001 Conference Proceedings

Abstract

We present a method for efficient global illumination computation in dynamic environments by taking advantage of temporal coherence of lighting distribution. The method is embedded in the framework of stochastic photon tracing and density estimation techniques. A locally operating energy-based error metric is used to prevent photon processing in the temporal domain for the scene regions in which lighting distribution changes rapidly. A perception-based error metric suitable for animation is used to keep noise inherent in stochastic methods below the sensitivity level of the human observer. As a result a perceptually-consistent quality across all animation frames is obtained. Furthermore, the computation cost is reduced compared to the traditional approaches operating solely in the spatial domain.




Paper

Download the paper in PDF [0.4 MB] or Gzipped Postscript (ps.gz) [3.5 MB].
Copyright ACM 2001.

Animation examples for the room and atrium scenes:

Room

Spatial filtering: off
On average the same number of photons per frame was used. In the case of 845,000 the same number of photons collected in the spatial domain was used as in the case of perception-guided approach for the temporal domain.

Atrium

Spatial filtering: on
On average the same number of photons per frame was used. (mov) is the QuickTime format with JPEG compression and half size.
(mpg) is the MPEG-1 format and half size.