Speaker: Christian Freude (ICGA)
This thesis proposes extensions for the Separable Subsurface Scattering algorithm to support arbitrary materials.
Four separable (rank-1) kernel models for the approximation of physically based diffuse reflectance profiles are presented. Each model offers different approximation quality and controllability. The first two models are based on singular value decomposition and a custom analytic pre-integration scheme. They enable fast deterministic kernel computation and provide fixed-quality solutions. Two additional parametrized models are based on automatic and manual optimization and provide more control over the approximation quality but are more time-consuming to generate. Higher rank approximations can be computed using the approach based on singular value decomposition. All four kernel models are used to compute approximations for physically measured diffuse reflectance profiles of different materials and tested using several special-case irradiance signals and complex proof-of-concept scenes.
The results are compared to the state of the art in real-time rendering of subsurface scattering, showing comparable approximation quality at lower computational cost. The proposed extensions enable rendering of physically based subsurface scattering for arbitrary materials and dynamic scenes in real time. Furthermore, this separable approach is already in use for SSS rendering in AAA games.