Boosting monte carlo rendering by ray histogram fusion

In this article we have introduced RHF, an adaptive filtering scheme that accelerates Monte Carlo renderers. In the proposed approach, each pixel in the image is characterized by the collection of rays that reach its surface. The proposed filter uses a distance based on the sample color distribution of each pixel, to decide whether two pixels can share their samples. This permits to boost the performance of a Monte Carlo render by reusing samples without introducing significant bias. We have presented several experiments showing that RHF achieves artifact-free high-quality noise reduction on a variety of scenes, and is able to cope with multiple simultaneous effects. The method is not only capable of removing high-frequency noise: thanks to its natural multiscale design, it can also successfully remove lowfrequency noise. The proposed method can be easily extended to process animated sequences. The method is independent of the rendering system and can be applied to samples generated by different methods, such as pure Monte Carlo path tracing or photon mapping with final gathering. It could also be potentially applied to postprocess other methods that resynthesize samples using information from the scene, like the one recently proposed by Lehtinen et al. [2012]. An advantage of the proposed filter is that its time and memory complexities do not depend on the number of input samples, and scale linearly with the image size. Finally, since a direct output of our method is the number of similar pixels for each given pixel, a decision on where to distribute new samples can be adopted. Thismay lead to an adaptive rendering version of the proposed filtering approach. © 2014 ACM 0730-0301/2014/01-ART3 15.00.