New Microfacet Based Reflection Models
Reflection models (or BRDFs, which stands for Bidirectional Reflectance
Distribution Function) are at the heart of computer graphics. This is what
determines the perception of what material every object in a computer generated
image is made of. There are several strategies to obtain this reflection
information.
First, one can measure the BRDF. This, of course, is the most
accurate and general (at least in theory) way, provided one has the equipment
and material sample available. Short of this, a model of reflectance can
be constructed from the first principles, based on physics of light-surface
interaction. The result of this is usually rather complex. Finally, one
can do what most people do - use some popular empirical model, such as
Phong model. These work surprisingly well given the simplicity but usually
handle a rather restricted class of surfaces and lack some desirable properties
needed to describe the true reflectance correctly.
In this project we explore a class of models which is somewhere
in between empirical models and truly physics based ones. These are microfacet
models. We were able to make them much more general and easier to use while
preserving their nice mathematical properties. The result is that we can
generate a whole family of new reflection models for surfaces with non-trivial
properties. The paper below describes the details.
From ACM SIGGRAPH 2000:
Based on this work, we were also able to create a new Phong-like
reflection model which has better energy behavior, accounts for Fresnel
effects and allows anisotropy. An attempt was made to preserve Phong parameters
(since most graphics people are quite familiar with what it means to have,
say, n=100 Phong model) and make this model as simple and efficient as
possible to allow a hardware implementation some day. Who knows, maybe
it will become as popular as the original Phong model. A short paper concentrating
mostly on how to implement the model is below.
From the Journal of Graphics Tools:
A Tech. Report UUCS-00-014 giving more justification to the model
and describing its connection to microfacets can be found here
(2.0 Mb).