Cost-Driven Multiple Importance Sampling for Monte-Carlo Rendering

The global illumination or transport problems can also be considered as a sequence of integrals, while its MonteCarlo solutions as different sampling techniques. Multiple importance sampling takes advantage of different sampling strategies and combines the results obtained with them. In this paper we propose the combination of very different global illumination algorithms in a way that their strengths can be preserved. To do this, we generalize the fundamental theory of multiple importance sampling for sequences of integrals and also take into account the computational cost associated with individual sampling techniques. The theoretical results are used to combine bi-directional path tracing and ray-bundles based stochastic iteration.