Theoretical Framework And Physical Measurements Of Surface and Subsurface Light Scattering From Material Surfaces.

PhD thesis, Cornell University, May 2005.

This thesis aims to improve fidelity of realistic synthesized images by providing practical solutions for simulating the surface and subsurface light scattering phenomena. The solutions achieve a good balance between the physical correctness and the computational cost, and are verified against light scattering measurements carried out as part of the thesis. The proposed theoretical framework includes:

* A BRDF model for simulating first surface reflections

* A BRDF model for simulating local subsurface scattering

* A hybrid method for simulating volumetric subsurface scattering

The first BRDF model accounts for first-surface reflections. The model combines highly-efficient wave optics components and rigorous empirical components. Consequently, the accuracy and generality of the model are comparable to those of the wave-optics model; while its computational cost is much lower. The model correctly predicts various light scattering phenomena and applies for to a wide range of materials and surface finishes.

The second BRDF model accounts for local subsurface scattering that shows no volumetric effect. The model describes the non-directional and directional subsurface scattering with physically-plausible mathematical constructions. The angular dependence of transmission and the directionality of the subsurface light transport are taken into consideration.

Both BRDF models compare favorably against extensive, detailed reflectance measurements that were carried out as part of this thesis. The models are analytic and suitable for practical Computer Graphics applications. Benchmark timings are comparable with that of current less comprehensive models (Lafortune, Ward, and Cook-Torrance models).

The proposed hybrid method is a flexible approach that efficiently simulates the appearance of a wide range of participating media, which are not well handled by currently available methods. The method combines the Monte Carlo technique and the dipole diffuse approximation. The first several scattering events inside the volume are critical for rendering, especially for highly-curved or optically-thin materials, and can only be well simulated by a pure Monte Carlo method. The contribution of the subsequent scattering events, which may not always exist, can be approximated by the dipole diffusion approximation with acceptable accuracy. While the accuracy of the hybrid method is comparable to a pure Monte Carlo method, its efficiency is much higher.