UC San Diego

Monte Carlo integration is a technique that numerically approximates a definite integral using random samples. Existing global illumination algorithms leverage this technique to estimate realistic lighting of 3D scenes. To half that estimation error, four times as many samples are needed. This thesis proposes an automated practical way of analyzing a given 3D scene and generate a modest amount of samples that dramatically reduces the estimation error. The results demonstrate that the estimation improved by several orders of magnitude despite using less samples