These Incredible Facial Scans Capture Detail Down to the Skin Cell

Rendering CGI faces that look close to real is hard, but we're starting to see hardware that can pull it off. Then, the problem becomes the source material; you've got to have models that look good enough to pass for real. For that, a new kind of facial scanning is going to come in handy, one that can get detail all the way down to each skin cell.


A team of researchers led by led by Abhijeet Ghosh of Imperial College London and Paul Debevec of the University of Southern California have developed a process of scanning human faces at such a high level of detail that the reproductions are effectively perfect. No pore goes unnoticed.

The process works thanks to extremely detailed application of light during the capture process. As New Scientist explains:

Using a specially developed lighting system and camera, they photographed samples of skin from people's chins, cheeks and foreheads at a resolution of about 10 micrometres, so that each skin cell was spread across roughly three pixels. They then used the images to create a 3D model of skin and applied their light reflection technique to it. The result was CGI skin complete with minute structures like pores and microscopic wrinkles. Finally, they fed the CGI images to an algorithm that extended them to fill in an entire CGI face.

Earlier versions of this same tech have been put to commercial use before. In fact, digital effects company Weta Digital used some of these same techniques in the production of Avatar. But many of the finer details that helped get the Na'vi across uncanny valley—like moles and wrinkles—are were added by hand. The latest and greatest version of this capture tech can pull that stuff off automatically, making it waaaay more practical and less labor intensive.

The crew at USC is already working with games publisher Activision to try to find a way to bring these sort of high quality faces to games as soon as possible, though it'll probably be a while before consumer tech can handle it. We're close to finally getting across the uncanny valley though. The porn can't be very far behind. [New Scientist]




The underlying premise of this story seems to be that there is a technique to producing extremely realistic looking people in games and film effect. This technique has to do with going to any almost microscopic level, and thats it. The notion that if we get closer and closer to the small details we can create something hyper real has long since ended. This is the brute force technique, we cant create truth through brute force, only imagination and experimentation can advance the ball.

To get hyper realistic humans many things are going on at once and no level of microscopic scanning will change this. There is no one answer or technique, I would even say that microscopic is the wrong direction if you want to gain greater realism in the human form. Hair and fur was a major breakthrough in effect this too was microscopic and none of it was a result of scanning it was about thoughtful analysis of the real world and great software writers experimenting for years. Did I mention specularity, mapping, rendering, lighting and motion.

While this new approach is somewhat different then past approaches it is very similar to several techniques that have been used for some time. Interestingly it leaves out something I have though would be a key to maximum realism, seeing below the skin. When we look at skin we see into it, we see below the surface, this effects texture and color. While this might seem like I am talking about a microscopic examination of each actor I am not, I am talking about using the technique that was used for hair and fur, a global rules based system which can be tweaked to create a look.

To me, the real give away to computer generated people is the way the people move. The singular talent for understanding and executing motion is everything. So much of how we absorb the world around us is in the periphery. We might look at someone walking towards us and think we are focused on a single part of that person, we are not, we are relaxed and viewing the amalgam of things not a single thing, computer generated people give it away on that level not the microscopic.

What is interesting about the work of this story is the notion of changing the face from a surface that must conform to a human look to a surface that is structure, with structure you can have much more control over movement and therefore a more realistic synthetic person, forget about the microscopic its neive.