Graphics Reference
In-Depth Information
based on the half-angle
θ
h
clearly produce more accurate approximations across
all the different surfaces.
8.2 The Reflectance Field
The methods described above allow for the recovery of BRDFs from photographs
of surface samples under controlled lighting conditions. It is assumed that the
pixel values of the captured images record the radiance directly reflected off the
surface from the single illumination source. That is, global illumination effects
such as indirect lighting are ignored. This is a reasonable assumption, because the
photography is done in a closed laboratory environment with black matte walls
or curtains that absorb stray light. The only light source is the surface illuminant.
There is some stray light in the environment: interreflections between the surface
sample and the camera and gantry device are probably the most significant source
of this. But painting everything black minimizes these effects. Recovering BRDF
data from photographs in an environment where there is significant global illu-
mination is an entirely different problem—global illumination effects have to be
separated from the direct reflection before the BRDF values can be obtained. This
section introduces some approaches to performing this separation, and describes
some other uses for the separation methods.
8.2.1 Inverse Global Illumination
Global illumination comes from three elements: the geometry of the scene, the
reflectance properties of the scene objects, and the light sources in the scene.
Methods described in Chapter 2 can be used to compute the GI solution for
the scene, which gives the outgoing radiance at each point on each surface. The
recovery of reflectance given the surface radiance along with the geometry and
lighting of a scene is a formidable inverse problem. This is one form of an
inverse
global illumination
problem. Figure 8.21 shows a schematic representation of the
conceptual process. The paper entitled “Inverse Global Illumination” by Yizhou
Yu, Paul E. Debevec, Jitendra Malik, and Tim Hawkins [Yu et al. 99], presents a
method for approximating inverse GI that allows for the recovery of BRDF data
from photographs in a general environment.
The method described in the paper attempts to reconstruct an approximate
BRDF model for surfaces in a photographed scene. The dense sampling over
many light and view directions used for more precise BRDF recovery methods
is specifically avoided—the approach uses a minimum number of photographs.
As illustrated in Figure 8.22, the angles the light and camera positions make with
