UC San Diego
Towards realistic image synthesis of scattering materials
- Author(s): Donner, Craig Steven
- et al.
This dissertation focuses on developing accurate yet efficient shading methods for realistic image synthesis of scattering, or translucent, materials. Translucent materials are characterized by a soft "glow" when backlit, and are often mistaken for purely diffuse materials when front-lit. This appearance is caused by the penetration and subsurface scattering of light. Accurately and efficiently simulating light transport in these materials is a challenging problem in computer graphics. In this dissertation, we derive the popular diffusion dipole method widely used to simulate light transport in highly scattering materials. We analyze its strengths and weaknesses, along with the specific approximations made along its derivation. We then introduce the multipole method, an efficient and accurate technique that simulates the transport of light in thin translucent slabs. It uses multiple dipoles mirrored about the boundaries of the slab to approximate boundary conditions on the internal radiance. We extend the multipole to model the inter- scattering of light between diffusing layers using a novel frequency domain technique similar to Kubelka-Munk theory. The new model accurately predicts the reflectance and transmittance of arbitrary systems of diffusing layers. To show the accuracy and validity of the model, we compare results of the multipole with those of Monte Carlo simulation, as well as data taken from the optics literature, and provide various photorealistic rendered images of materials such as paint, paper, and human skin. As an example of a complete shading model, we also develop a new practical spectral model for shading human skin. We model skin as a two-layer translucent material using the multipole method. The model is driven by only four physiological parameters. These parameters control the amount of oil, melanin, and hemoglobin in the skin, which make it possible to match specific skin types, and also give a more intuitive control to drive the multipole. As validation, we compare simulated reflectance spectra of skin to actual measured data, and provide several photorealistic renderings of skin under various lighting conditions