www.karsten-schwenk.de

Practical Noise Reduction for Progressive Stochastic Ray Tracing with Perceptual Control

Schwenk, Karsten; Kuijper, Arjan; Behr, Johannes; Fellner, Dieter W.:
Practical Noise Reduction for Progressive Stochastic Ray Tracing with Perceptual Control.
Computer Graphics and Applications, 2012. To appear.
publisher's site
preprint.pdf
supplemental.zip

We present a method to reduce noise in stochastic ray tracing that is especially tailored to interactive progressive rendering. High-variance light paths are accumulated in a separate buffer, which is filtered by a high-quality edge preserving filter. Then a combination of the noisy unfiltered samples and the less noisy (but biased) filtered samples is added to the low-variance samples in order to form the final image. A novel per-pixel blending operator combines both contributions in a way that respects a user-defined threshold on perceived noise. Our method is able to provide fast, reliable previews, even in the presence of complex features like specular surfaces and high-frequency textures. At the same time it is consistent in the sense that the bias due to filtering vanishes in the limit.

viper_filter.avi
Small video of an early version of the algorithm. The compression makes it look a bit blurrier than it really is, but I think the point gets across.