This is not a computer-animated blond, but she may soon be watching one on that screen. Credit: Robert Kyllo
In the world of computer animation, creating King Kong is a snap; it’s his pretty blond lady friend that causes all the trouble.
However, a team of Cornell researchers recently developed an algorithm that will allow computer animators in Hollywood to quickly render realistic-looking blond locks.
As more and more movies make use of computer-generated images, a cadre of scientists is helping Hollywood make its computerized characters look real. Hair, it turns out, is one of the most difficult things for computer animators to portray.
A typical head has about 100,000 strands of hair, each of which is distinguishable from those located nearby. This makes determining exactly how light interacts with those strands something of a knotty problem.
Dark hair, like that of Kong, is relatively easy to render convincingly on screen. A ray of light traveling out from a light source is absorbed quickly by dark hair, so animators typically only have to determine how to represent the small percentage of light that is reflected back.
Blond hair absorbs much less light—most travels through the hair and scatters off multiple strands.
“Rather than coming back to the viewer, you end up with light that bounces around many hairs,” said Stephen Marschner, a computer scientist at Cornell and co-developer of the new algorithm.
Capturing this scattering pattern requires complex calculations and a laborious method known as path-tracing, which can take days for a computer to perform. Animators typically rely on a shortcut: they assume that hair is opaque and that light hitting it simply bounces back.
“That works fine for black hair,” Marschner said. “But for lighter colored hair, you get it completely wrong.”
Marschner and his colleague Jonathan Moon decided to adapt a method known as photon mapping to provide animators with a program that would generate realistic-looking blond hair.
Photon mapping involves determining how the light in a given scene interacts with an object. The method developed by the Cornell team begins by distributing millions of particles of light—or photons—throughout a scene by shooting each one at the total volume of hair and following its interactions. Each time a photon encounters a hair, the algorithm determines whether it’s absorbed, reflected or travels through the strand. The computer stores each photon’s path, recording information about its position and direction at given time intervals. This data is then compiled into a map that estimates where each of the photons in the scene are at a particular point in time.
With the photon map complete, the computer can then consider each pixel of hair. The algorithm uses the stored photon map to make a density estimate for each pixel of hair visible in the image, evaluating the direction and position of photons near the point in question. The result is an estimate of the radiance arriving at the hair in that part of the image. Doing this for each component of the image yields a realistic simulation of a full head of blond hair.
“There’s been a lot of work on human skin,” said Henrik Jensen, a computer graphics specialist at the University of California, San Diego. “Of course, humans—and many other creatures, [such as] King Kong—need to have hair, as well.”
“It builds nicely on the previous work on hair,” he continued.
Marschner and Moon’s photon mapping technique is far faster than the path-tracing method. It produces an image in a few hours that formerly would have taken days to generate.
Hollywood has already expressed interest in the new method, said Marschner, winner of a technical achievement award from the Academy of Motion Picture Arts and Sciences for work enabling animators to create realistic-looking skin for the character Gollum in The Lord of the Rings.
Despite such progress, mastering blond hair doesn’t mean we’ll be mistaking a computer-animated person for a real one anytime soon.
As Marschner said, “There are a lot of different problems involved in synthetic humans.”
Originally published August 4, 2006
