Additional Progress Towards the Unification of Microfacet and Microflake Theories
EGSR 2016 (Experimental Ideas & Implementations track)
Current computer graphics materials suffer from darkening effects at high roughness values. Darkening can be very disturbing for certain types of rough materials such as frosted glass, as illustrated below (notice how the material becomes darker as roughness increases; matching the material of the photography is impossible without code hacking). In this work, we are interested in supporting such rough materials in a more realistic way.
Current computer graphics materials are derived from microfacet theory. Microfacet theory explains the behavior of a material by considering a surface composed of microscopic mirrors that can deviate incident light rays. The average deviations define the material, whose roughness is directly related to the way the mirrors are statistically oriented. For instance, if the mirrors are mostly aligned towards similar directions, then the material is smooth. Conversely, if the mirrors are mostly aligned towards different directions, then the material is rough.
Microfacet materials have been very successful due to their computational efficiency and artistic control, but suffer from darkening effects at high roughness values, as we saw earlier. Darkening comes from the fact that microfacet theory only accounts for one light deviation within the material. For rough materials however, light is expected to undergo multiple deviations within the material, because the probability of interacting with more than one mirror increases. In order to create more realistic rough materials, we need to extend microfacet theory to account for multiple light deviations.
We extend microfacet theory by treating rough surfaces as participating media. We show that microscopic mirrors suspended in thin air deviate light the same way as if they were assembled into a surface. Thanks to this equivalence, we can account for multiple light deviations and recover the light that was ignored by the previous model. Below is a frosted glass material generated with our model; notice how the material closely matches that of the photography.
Extending the theory is only a first step towards better looking materials: currently, our model is much more computationally expensive than the previous microfacet materials. This is why we are investigating ways to reduce this gap in order to bring more realistic rough materials to computer graphics.