Photon mapping requires the reconstruction of the lighting using virtual photons in a computer generated image. In traditional approaches, there is a trade-off between variance (noisy images) and bias (blurry images). This can be mitigated by using an order of magnitude more photons, but doing this adds to the time to reconstruct the light, and still suffers from bias versus variance trade-off.
The photon relaxing method addresses the problem of bias-variance trade-off and describes a totally new and counter-intuitive approach to reducing such errors from the underlying photon distribution of a pixilated graphic by repositioning the photon as a function of its proximity to its nearest neighbours making a judgement on the size and direction of positional error based on where its closest neighbours are located. This algorithm combines point relaxation processes with feature detection that analyses the photon distribution of a graphic and isolates points which lie near edges, boundaries and other important visual cues to keep graphic outlines crisp. The method produces dramatically more realistic images with fewer photons in the estimate, and therefore speeds up rendering time, and greatly reduces blurring and noise. It also reduces memory overhead since scenes can be rendered using fewer stored photons.
This rendering system reduces hours spent positioning lights to work out light distribution to be used in computer animated films and games to minuets. A quicker rendering time will also reduce the amount of internal memory storage required during the production of computer animated films. Manufacturers of games, films and architecture applications will ultimately benefit from this interactive technology.
Inventors: Mark Jones & Ben Spencer