
Strange as it sounds when discussing an 8-metre-tall gorilla atop the Empire State Building, you’d be hard-pressed to discern from looks alone that the eponymous hero of the recent movie isn’t real.
His black fur glints in the sunset and ripples in perfect sync with his movements, the product of software that represents every single hair as a translucent cylinder covered in tiny scales. Simulated light rays are scattered, reflected and refracted almost exactly as they are by real hair. This sheen helped King Kong win an Oscar for visual effects in 2005.
Now its creators say the software might have a radically different use: because the hair it simulates is so realistic, it could be used as a testbed for future hair products. Companies could use the model to represent human hair and then add a variety of simulated dyes and conditioners. That would enable them to explore how those products make hair look, without going to the expense of having to manufacture them all first. Steve Marschner of Cornell University in Ithaca, New York, is a member of the team that created the software and is currently working with Unilever, which makes hair-care products.
Advertisement
“Computer graphics has now reached a point of sophistication where it can be used beyond the fields of gaming and films,” says computer scientist Henrik Wann Jensen at the University of California, San Diego, another of the team that developed the hair used on Kong. Indeed, that software is just one example of a range of computer graphics which have become so photorealistic that their applications have surpassed simple entertainment.
The refinements to reach this stage have been extensive. For decades, animators have been able to simulate how light beams bounce off objects, but animated characters in games and movies were traditionally made to look realistic thanks to the talents of artists who touch-up the resulting images until they look convincing. This wasn’t enough to create a realistic simulation, however: while an animated character may look plausible in a snapshot, a lack of accurate built-in physics meant that it appeared unnatural in moving scenes.
If creating realistic animation in this way is difficult, it is near impossible to make touched-up animations look right when they are merged with live action in films like King Kong. In this instance, the ape needed to look like he was illuminated by the same light sources as the characters in the live action shots, all of whom are in motion. To tackle the problem, the movie’s creators turned to models that simulate the way that light propagates through different materials in much greater detail. Unlike artistic impressions, the result is models that look realistic no matter what situation they are placed in.
Hair is a great example. Earlier models treated each hair as an opaque cylinder, causing the light to propagate in a way that made the hair look dull and lifeless. But Jensen, together with Marschner and Pat Hanrahan of Stanford University in California, discovered that the key to making hair look realistic was to create a model that was closer to its actual structure – translucent and covered in tiny scales (èƵ, 2 August 2003, p 19). Also critical was their discovery that the cross-section of hair is elliptical, which is important in creating glints. By creating models that mimicked this structure and simulating how that would affect light shining on it, they created the realistic fur that adorned Kong. “We keep pushing the limit – taking more information into account,” says Jensen.
Healthy glow
The rendering of skin, historically one of the most difficult targets for physics-based models, has also come on in leaps and bounds. In 2001 Jensen, Marschner and Hanrahan found that, as with hair, the key to rendering skin realistically is to recognise that it is , not opaque. In the new software, light not only bounces straight off the skin but also enters the skin and moves beneath the surface before being reflected back out. This insight gave the skin a realistic glow. This approach was used to create ‘s skin in The Lord of the Rings trilogy. The ability to model translucency also allowed the team to tackle another murky problem in computer graphics – milk (see “Got Milk?”).
Jensen is currently making the skin model even more realistic. In 2006, he came out with an improved version that into two layers: an epidermis layer containing models of two types of melanin and a dermis containing haemoglobin. “Without this, there’s a uniformity to the skin that may not be quite right. Things start to look a little bit like wax,” he explains.
Jensen now plans to model the skin in even greater detail, incorporating individual cells and fibres within the layers. He says these models could be used by the cosmetics industry to create more natural-looking foundation. Researchers could put the virtual foundation on the simulated skin to estimate that foundation’s light-scattering properties and see how it affects the natural translucent look.
The skincare industry is already showing interest in skin models. Computer scientist Shree Nayar of Columbia University in New York has studied models of with Takanori Igarashi from Kao Corporation, a Japanese company that makes skincare products. Meanwhile, Kao, Procter & Gamble and Unilever have all expressed interest in the skin model built by Jensen, Hanrahan and Marschner.
The model might also have healthcare applications. Jensen says it mimics real skin so closely that a future version could be used to simulate how far light of particular intensities propagates through a cancer patient’s skin. This could be used to determine how big a dose of tumour-zapping light or laser therapy they need.
Got Milk?
In Shrek, some say the most difficult shot to produce was that of a small glass of milk. By the time came out in 2004, vastly improved software for rendering milk meant that the guards in the sequel went crazy for the stuff, even going so far as dumping boiling milk on a walking gingerbread man.
Milk was previously difficult to model realistically because it is translucent. In the first Shrek, it was modelled as an opaque fluid, which meant the light bounced straight off its surface, making it look like paint.
To build a realistic model of milk, in 2001, Henrik Wann Jensen at the University of California, San Diego, and colleagues added reflections from light scattering beneath the milk’s surface. They used a technique that was later used to make Gollum’s skin look eerily realistic in The Lord of the Rings trilogy. Now, insights gained during this process are being put to work in the dairy industry, in the name of quality control.
To model just how light moves under the surface of a substance, Jensen specifies the substance’s ability to scatter, absorb, refract and spread light. He deduces what values each property should have for a given substance by shining a spot of light onto a sample and measuring how the light intensity fades from the centre of the spot. Software then uses those properties to create a realistic model of the light moving and scattering beneath the surface.
Now Flemming Møller, a researcher at Danish food-ingredient company Danisco, is borrowing Jensen’s technique to help determine particle sizes in drinking yogurt and to measure the size of air bubbles and ice crystals in ice cream – important for quality control and standardisation. Like Jensen, he shines a spot of laser light on the yogurt or ice cream. As he has already correlated how the resulting pattern varies with particle and air bubble size, he can determine them from the shape of the spot. This allows Møller to test the products’ quality without having to sample the food invasively, something that always carries a risk of contamination. It also removes the need to dilute the samples, which is necessary for standard light-based tests.
The technique is not used routinely at Danisco but Møller hopes it will be become widespread. “This work has been an eye-opener,” he says. “I thought that computer graphics were very simple – you sit down and it’s a lot of nerds. I was very surprised that there was a lot of science behind it.”
Compliments aside, Jensen has since updated the milk model so that it can be programmed to vary the sub-surface scattering and reflection according to the of the milk. The primary light-scattering particles in skimmed milk are clumps of protein, but whole milk also contains fat globules, which are larger than the protein clumps. Jensen’s model uses this to work out how to vary the way milk looks according to the fat and protein composition. He found that skimmed milk looks bluish, because protein molecules scatter blue light preferentially and whole milk looks white, because fat globules scatter all frequencies equally.
He can also reverse the process to determine the fat and protein content of a sample of milk – and therefore the type of milk – just by shining light on it. He does this by running multiple milk simulations, tweaking the fat and protein content with each run until the optical properties of the simulated milk – and therefore the fat and protein content – match that of the real thing. Møller hopes to use the same technique to more precisely determine particle size in a sample.
Jensen believes that such models will have other applications. By measuring how pollutants, plankton, minerals and algae affect the optical properties of seawater, a model similar to the milk model could be used to monitor and interpret changes in the oceans, he says. And a model of the atmosphere might allow changes in its composition to be tracked.