Real-time kd-tree based importance sampling of environment maps

We present a new real-time importance sampling algorithm for environment maps. Our method is based on representing environment maps using kd-tree structures, and generating samples with a single data lookup. An efficient algorithm has been developed for real-time image-based lighting applications. In this paper, we compared our algorithm with Inversion method [Fishman 1996]. We show that our proposed algorithm provides compactness and speedup as compared to Inversion method. Based on a number of rendered images, we have demonstrated that in a fixed time frame the proposed algorithm produces images with a lower noise than that of the Inversion method. We also demonstrate that our algorithm can successfully represent a wide range of material types.