If you've seen the film The Social Network, then you might have wondered about the identical Winklevoss twins - were a real-life pair of twins cast for the roles, or was it a bit of Hollywood magic? Well, it was magic. Although two different actors' bodies were used, their faces both belonged to actor Armie Hammer. After the movie was shot, the body double's face was digitally replaced with Armie's. While such computer-enabled face-swapping trickery has so far been available only to feature film-makers with deep pockets, that could be about to change, thanks to research being conducted at Harvard University.
Computer scientist Kevin Dale has created a digital "face transplant" system, that reportedly works with a single camera, simple lighting, and a regular desktop computer.
The process begins with footage of a supplying "source" actor, and a receiving "target" actor - both of which can be talking. The software proceeds to create 3D models of both of their faces, then subtly alters the pose and speech of the target actor to match those of the source. Needless to say, it works best if both actors are in the same pose, in the same lighting, and saying the same thing.
On a frame-by-frame basis, the outline of the transplanted face is blended with the skin of the receiving actor - in this way, a flickering seam around the face is avoided when all the frames are run together. Additionally, the software chooses a location on both faces, both for cutting out the one and covering over the other, where the boundary between the two will be the least noticeable.
Ten seconds of finished footage takes about 20 minutes to render, with minimal input from the user.
Not only could Dale's system be used to put one actor's face on another's body, but it could also combine different parts of the same actor's face, shot in different takes of the same scene. If the director liked the way an actor's eyes looked in Take 2, for instance, those eyes could be blended into the actor's face from the otherwise-better Take 4.
Source: New Scientist