Cache de luminance pour l’éclairage global en présence de surfaces rugueuses

Computation of global illumination is an important means in the pursuit of photorealism of computer-generated images. It is of particular importance in scenes with glossy surfaces, since it conditions correct perception of glossy material characteristics. This dissertation focuses on efficient computation of global illumination in presence of glossy surfaces with low-frequency reflectance functions. Computing global illumination on those surfaces is difficult with pure Monte Carlo sampling because of the sheer number of rays required to get a good estimate of indirect illumination. We propose radiance caching, a ray tracing-based algorithm for computing indirect illumination on glossy surfaces with low-frequency reflectance functions. The algorithm takes advantage of the shading smoothness on these surfaces by interpolating indirect illumination from a set of sparsely distributed, cached values. We discuss all the components necessary for an implementation of radiance caching: (1) an automatic choice of surfaces on which radiance caching is used; (2) compact representation for the cached illumination by spherical harmonics; (3) efficient spherical harmonics rotation that makes interpolation possible; (4) computation of translational gradient for smooth interpolation; and (5) an adaptive criterion for distributing the cached illumination values on surfaces. Radiance caching delivers high quality images in a shorter time compared to existing ray tracing-based methods. It can be used for final gathering in two-pass global illumination methods or as a stand-alone global illumination algorithm.