This research effort is directed at developing improved global illumination algorithms for image synthesis. Monte Carlo techniques avoid artifacts due to the discretization of the environment required for radiosity methods and are favored when the geometry of the rendered objects are freeform surfaces rather than polygonal objects. The major problems with the Monte Carlo methods are the noisy characteristic of the solution and the high sample rate required to bring the noise down to an acceptable level. Importance sampling and stratified sampling are two methods that can significantly reduce the noise level and sample rate required for a high quality solution. Radiosity methods can be used to provide guidance for importance sampling. Model preprocessing can be used to locate known discontinuity curves on the model surfaces and to partition the surfaces into pieces that are expected to have smooth illumination for stratification. Trimmed spline representations can be used to represent the illumination functions over each surface.
