Graphics Reference
In-Depth Information
and interreflection and also subsurface scattering. A SVBRDF normally does not
include these effects; they are computed during the rendering process.
Before BTF synthesis was conceived, there were attempts at capturing vari-
ation in reflectance with changing position within a small area using SVBRDFs.
The paper “Efficient Rendering of Spatial Bidirectional Reflectance Distribution
Functions” by David K. Mcallister, Anselmo Lastra, and Wolfgang Heidrich is
said to be the representative work in the subject [Mcallister et al. 02]. The method
described in this paper captures images of a small area of a surface sample with
varying lighting and viewing directions (in a manner similar to the way BTF data
is acquired). A set of BRDF values is obtained from these images, and the Lafor-
tune model is fit to each point on the sample surface. The authors describe a
hardware implementation that stores the Lafortune parameters in textures, and
employs a GPU shader to perform the rendering. The method presented in this
paper was a forerunner to later developments in hardware BTF rendering.
9.2 BTF Synthesis
A BTF represents the interaction of the light mesostructure of a surface on a small
area of the surface. This small area is assumed to be a rectangular planar patch.
The original BTF paper by Dana et al. concentrated on the measurement process
and the application to BRDF measurements. The authors suggested how BTFs
could be used in image synthesis, by tiling the BTF of the small sample across
larger surface in an environment. But they also note some potential problems. One
is the issue of seams between adjoining tiles, which would have to be covered or
smoothed in some visually plausible way. There is also the potential problem of
the repeated pattern being noticed by human observers, and this would affect the
realism of the rendering. Another is the problem of interpolating the BTF data for
arbitrary lighting and view directions. This kind of interpolation has come to be
known as
BTF synthesis
, as it involves synthesizing a new image for an arbitrary
viewpoint. (The term “view synthesis” is used elsewhere, e.g., in the interpolation
of light stage images.) In the case of BTFs, proper synthesis involves more than
just interpolation; it requires recovery of the local mesostructure.
BTF synthesis can be viewed as a generalization of
texture synthesis
.The
earliest implementations of texture mapping used a small texture map that was
replicated across a surface. In order for this to work, the texture map has to be
constructed so that the seams between the copies are not very noticeable. It was
not long before researchers looked for ways of doing this automatically, which led
to the more general problem of nonuniformly replicating the content of a small
texture map into a larger map that
looks
like the same texture image. This is the
