Irradiance filtering for Monte Carlo ray tracing

Stochastic ray tracing algorithms generate photo-realistic images by simulating the global illumination. Typically a rather long computation time is required for decreasing the visible noise to an acceptable level. In this paper we propose a spatially variant low-pass filter for reducing this noise. We analyze the theoretical background of the method and present an efficient implementation that enables the use of a comparatively small number of samples while producing high quality images. Our algorithm can be used to accelerate path tracing and final gathering in photon mapping. We compare the method to irradiance caching and the results show that our algorithm renders images of similar or better quality up to five times faster.