Graphics Reference
In-Depth Information
from
A
j
to
P
i
and
A
j
to
C
k
. But because of the possibility of an off-specular peak,
a set of rays is traced in small cones around these two reflected directions. If the
average radiance in these cones are denoted by
Q
P
i
,
A
j
and
Q
C
k
,
A
j
, respectively,
then
Δ
S
can be approximated by
Δ
S
=
ρ
s
(
A
j
)
(
Q
P
i
A
j
−
Q
C
k
A
j
)
,
where
ρ
s
,
A
j
is the average specular reflectance of patch
A
j
.
The algorithm described in the paper actually uses a hierarchical approach,
and applies the iterative algorithm on a set of linked patches determined in ad-
vance by examining specular highlights in the images. Choosing a set of linked
patches that properly represent the specular reflection is critical to the success of
the algorithm. The average radiance values in the cones come from the direct
recorded radiance values at the ray intersections. This amounts to a “one-bounce”
approximation to
S
. The authors note that path tracing could be employed to
obtain a better approximation, but the one-bounce approximation is apparently
sufficient.
Using a judiciously chosen set of sample patches (points), BRDF parameters
can be recovered from even a fairly small number of captured HDR photographs.
Because the BRDF parameters are recovered separately for each patch, spatially
varying BRDFs can be captured as well. However, at least one of the images must
capture a specular highlight of each patch for this method to work. Furthermore,
since there is a need to accurately describe the interaction of light between patches
to create the radiosity equation, the geometric model of the scene needs to be
precise. In practice, these constraints make it difficult to apply the algorithm
in a general environment. But if the geometry of an environment is accurately
reconstructed, the method can be very useful.
Δ
8.2.2 Reflectance Field
The
light field
(or
lumigraph
) described in Chapter 5 records the 5D
plenoptic
function
, the radiance at each point in space and each direction. The image of an
object contained within the light field from any viewpoint can be reconstructed as
a kind of slice of the light field. The light field representation is independent of
the geometry of the object (although the detail of a light field image is limited by
the sampling resolution of the representation). An analogous construct for surface
reflectance was proposed in the paper entitled ”Acquiring the Reflectance Field
of a Human Face” by Paul E. Debevec, Tim Hawkins, Chris Tchou, Haarm-Pieter
Duiker, Westley Sarokin, and Mark Sagar [Debevec et al. 00]. The
reflectance
field
, as the construct is called, was developed for the purpose of representing
light reflection from a human face.
