The photon map is a novel global illumination algorithm developed by Henrik Wann Jensen. It differs from standard ray tracing paradigms in that it is a forward ray tracer, i.e. tracing rays from the light sources along the surfaces towards the viewer. This contrasts with the usual backward methodology of tracing rays from the viewer along the surfaces back to the sources. The photon map has the primary advantage of handling the phenomenon known as caustics efficiently -- highlights on diffuse surfaces from specular reflection, which backward ray tracers have difficulty with due to the stratified sampling scheme which is often used to sample illumination on diffuse surfaces.
The photon map is based on a Monte Carlo (light) particle transport simulation which couples to a backward raytracer to effectively yield a bidirectional raytracer which delivers a full global illumination solution. Supplementing an existing backward raytracer with the photon map is therefore relatively straightforward. The mechanisms by which the coupling takes place differ for various forward raytracers. In the case of the photon map, the results of the forward pass are stored in a kd-tree data structure, which allows efficient nearest neighbour lookups during the backward pass, thus coupling the two passes.
In the context of a daylighting study, the photon map's ability to efficiently and accurately handle caustics is particularly relevant when daylight systems are involved in the proposal. These systems include light shelves, blinds, and lightpipes, and typically consist of specular materials which redirect daylight for glare reduction or increased autonomy from artificial lighting. These redirecting effects are effectively caustics, and are difficult to simulate accurately with a backward raytracer.
The easy part of the thesis involves implementing the photon map in the RADIANCE framework. RADIANCE is the backward raytracer in this case, and acts as the testbed along with its many auxiliary programs. The major difficulty in this work arises from the fact that the algorithm must be validated with regard to accuracy. Analytically checking simple case studies is one way, but its scope is very limited due to the complexity of the equations involved. The bulk of the work (and time) has been invested in an experimental validation using a scale model with different material combinations, and this is where things can get nasty. Apart from dealing with the spatial distribution of the light source (we restrict ourselves to an artificial source), we also have to deal with the material BRDF (bidirectional reflection distribution function), which we obtained from measurements using ISE's goniophotometer, one of only a handful on the planet.
Tackling these difficulties has become the focus of the thesis. Very few experimental validations have been undertaken in the computer graphics field. Ultimately, the results should justify the application of physically based rendering algorithms like RADIANCE and the photon map to "real world'' problems like lighting design.