Graphics Reference
In-Depth Information
Figure 2.6
The principle of photon mapping. In a first phase, particles are shot from the light source
and surface intersections are recorded in a “photon map.” The second phase renders the
scene using the photon map to estimate illumination at surface points.
form of particle transport simulation starting from the light sources; the second
involves tracing rays from the viewpoint, with radiance calculated from the result
of the particle simulation. In the simulation phase, a collection of packets called
“photons” are fired in random directions from the light source following normal
light paths. At a surface intersection, a photon is either absorbed by the surface or
it loses some of its power and continues along another path, or possibly splits into
two photons. In either case, information about the collision, including the power
deposited at the surface, is recorded in the
photon map
. Eventually each photon is
absorbed at some surface. The second phase consists of basic ray tracing from the
viewpoint. At each ray intersection, the photon map is queried to determine the
illumination, much like a final gather in the radiosity method. Photon mapping is
a form of Monte Carlo simulation
For MCPT, it may seem more natural to consider only paths that start at the
light and end at the viewpoint. However, constructing such paths according to a
uniform random distribution is a significant challenge. One method for doing this
was presented in a 1997 SIGGRAPH paper by Eric Veach and Leonidas J. Guibas.
The method, known as
Metropolis light transport
(MLT), starts from a few sim-
ple light paths and develops many more through guided randomized mutations
of each path. The construction of paths is based on the
Metropolis
algorithm.
3
The MLT algorithm is significantly more efficient than ordinary MCPT in most
environments.
3
This algorithm is named after its primary developer Nicholas Metropolis, a physicist who pio-
neered the use of Monte Carlo methods. It is also known as the Metropolis-Hastings algorithm.
