Graphics Reference
In-Depth Information
a result, the hair looks too well “combed.” Nonetheless, it clearly shows how well
Marschner's model reproduces the appearance of real hair.
The first production use of the Marschner model was in the movie
King Kong
[Jackson 05]. The New Zealand digital visual effects company Weta Digital,
which did the visual effects for the movie, used Marschner's model for King
Kong's fur as well for all the “digital doubles” (CG characters that replace ac-
tual human actors). Rendering this human hair included the greatest challenge
yet for hair rendering: the blond hair of heroine Ann Darrow played by actress
Naomi Watts. Martin Hill and his colleagues at Weta implemented the Marschner
model as a RenderMan shader.
The preliminary results had the look of a “shampoo commercial”—the hair
looked too good. This was not such a problem for the human characters, but it
was less suitable for Kong himself. He was, after all, a wild gorilla. A series
of different implementations were required to get the desired look. In order to
produce a dusty appearance, an analytic cylindrical diffuse model was layered on
Marschner's model. In other cases, geometric debris was stuck between multiple
hair strands to form clumps. Glossy highlights were added to produce the appear-
ance of wetness. For shadowing within the fur, deep shadowmaps (Section 3.3.8)
were used in conjunction with volume occlusion maps, which work like ambient
occlusion maps in volumes. The amount of occlusion is computed for each point
in the fur in a reference pose, and this is stored as a 3D volume that can be ac-
cessed at render time to obtain the occlusion. The “bent normal” approach used
in the ambient occlusion map (Section 7.2.6) was also employed. Later, Weta
Digital developed a more physically accurate rendering pipeline that included the
effects of multiple scattering.
8.5 Multiple Scattering in Hair
The primary characteristic of light colored hair is its translucency. As described
in Chapter 4, multiple scattering is an important element of the appearance of
translucent objects, so accounting for multiple scattering is necessary to real-
istically render light colored hair. The direct and indirect separation shown in
Figure 8.50 affirms this. The Marschner model considers only reflection and
transmission in an individual hair fiber. A “head” of hair consists of many thou-
sands of hair fibers, and the interaction of light between them is not directly
handled by Marschner's reflectance model. The images in Figures 8.57 were
rendered by ray tracing a complex geometric model with each hair strand explic-
itly included as a geometric primitive. At an intersection with a hair fiber, the
Marschner reflectance model was used to compute the surface radiance. The ray
