Pixelcuts : Scalable Approximate Illumination from Many Point Lights

We present pixelcuts, a scalable algorithm to compute approximate low-frequency illumination from many point lights. Its running time is Õ(n+mk) where n is the number of pixels, m is the number of point lights, and k is a constant the user specifies. Our algorithm was inspired by the lightcuts approach, which runs in Õ(nk + m) [15]. Lightcuts builds a tree of point lights and, for each pixel, gathers contribution from O(k) light groups. On the contrary, pixelcuts builds several trees of pixels, and, for each light, scatters its contribution to O(k) pixel groups. We demonstrate that pixelcuts outperforms lightcuts when there are many more pixels than lights. Nevertheless, pixelcuts is limited to low-frequency illumination. It is also prone to structured noise if too few trees are used or if the pixels are divided into too few clusters.