Importance methods for photon mapping

Abstract

Photon mapping is a global illumination method widely used in realistic image synthesis. It is a Monte-Carlo Sampling based technique and noise is a potential problem in the rendered image. Tracing high number of photons reduces the noise but this is computationally expensive. Importance based methods provide a means to reduce the noise using a relatively low number of photons during tracing. These methods can be classiffed into two categories: importance sampling methods, used at the rendering phase to improve the sampling of nal gather ray directions, and visual importance methods, used at the photon tracing phase to send more photons in visible areas. In this work, we implement a number of importance based methods, and discuss their advantages and disadvantages. We present an importance sampling method that uses photon densities, and show how it reduces noise especially in scenes having non-uniform illumination. We provide a hybrid visual importance method and discuss its effectiveness. We also propose a photon power distribution method as an add-on to visual importance methods, which gives good results when used in images with highly-varying photon powers.