Graphics Reference
In-Depth Information
Figure 1.29.
Square light shape using Phong distribution.
1.6 Further Research
In the near future we would like to focus on enhancing the normalization part of
the specular integral, which currently, due to disk approximation (Section 1.3.4),
is lacking in visual quality when compared to the reference solution. We would
be especially interested in pursuing a good approximation for Blinn-Phong and
therefore other microfacet distributions with a similar base such as the recently
popular GGX.
In order to do so, we would need to find an ecient way of finding intersections
between rotated ellipses and disks or rectangles.
We also feel that the rectangular lights specular approximation, when using
nonradially symmetrical PDFs, needs more research. As of now, the current algo-
rithm for the most important point might result in severe artifacts at a rotation
angle of light that is orthogonal to the anisotropy direction of the PDF. A proper
search algorithm would require taking anisotropy direction into account. It is
worth noting that even in its current form, artifacts might be visually bearable
for rectangular shapes close to square (see Figures 1.29 and 1.30).
One proposed heuristic would be to treat the projected specular cone as an
elongated rectangle, oriented and extending in the direction of anisotropy. Such
shape change could be dynamic, calculated for specific BRDFs. Then, the most
important point for specular sampling would be the geometric center of the in-
tersection between the specular rectangle and the light shape. The intersection
area could also be used as
A
s
for normalization (see equation (1.20)).
