Graphics Reference
In-Depth Information
In basic Monte Carlo ray tracing, rays are shot from the viewpoint or camera.
When a ray hits a surface point of a caustic, finding the path to the light through
the refractive media relies on random paths eventually hitting the light—there
is no “shortcut” straight to the light as there is for the direct illumination. The
probability of hitting the light source depends on the size of the light. If the light
is large enough, it is likely that enough random paths will hit the source to capture
the caustic. However, the most visually interesting caustics come from small light
sources and therefore require a prohibitively large number of samples to capture
effectively.
A solution is to trace paths from both the light sources and the camera. But
bidirectional path tracing is complex, and is not particularly suited to rendering
caustics. Photon mapping traces paths in both directions, but in two separate
phases. The first phase is a particle simulation starting from the light sources,
which distributes the particles throughout the scene. The second phase uses ray
tracing from the viewpoint. The radiance at each surface intersection is computed
from the locations of the shot particles, much like the final gather in radiosity,
except that the photon map replaces the surface patches. The second phase is
the main rendering stage; the first is a preprocessing phase that is the essential
component of photon mapping.
In photon simulation, particles simulating the transport of light energy are
fired in random directions from the light sources and then are used to simulate the
interactions between light and surfaces in the scene. Particle intersections at the
surface are recorded for use in the rendering phase. The particles are called “pho-
tons” but they are not true photons in the sense of quantum optics; they merely
represent packets of radiant energy.
In the particle simulation phase each photon is given a fraction of the radiant
power (flux) of the light source from which it is emitted along with a direction
vector (
Figure 2.11
). The radiant flux of a photon is denoted by
ΔΦ
p
(
x
,
ω
p
)
.
(2.13)
Every time a photon interacts with a surface, it stores its incoming direction,
power, and position into a 3D map called the
photon map
. The photon may also
be reflected, or “split” by transmission, according to the physical properties of
light rays. Caustics are created by firing enough photons to make the photon map
dense enough for faithful representation of the caustic.
Photon mapping can be used to make MCRT more efficient in general, as it
provides an alternate form of irradiance caching. Basic irradiance caching only
records the direct irradiance from the first diffuse bounce (although path trac-
ing can be applied to compute secondary bounces). The photon map accounts
