University of Washington researchers have demonstrated software that allows them to (among other things) take the words and expressions of former President George W. Bush and have a 3D model of President Barack Obama act them out. The software uses old photos (a lot of them) and video to map faces, but what’s very impressive is the degree to which it can find parallel expressions.
The UW team is presenting a paper on December 16 at the International Conference on Computer Vision in Chile that will explain the face reconstruction, tracking, alignment, modeling, and puppeteering technology it has been developing over the past five years—and demos of the results are already here:
These animated representations are born from formulas that map 49 reference points on a human face, and then use existing photography to chart the changes in each point depending on facial expression.
However, as the videos demonstrate, the common traits between faces can be exploited so that the words of one person can be applied in a way to seemingly come from someone else’s mouth:
There are techniques already in use that can generate avatars of real people for virtual environments—but the goal of the UW researchers, as they have publically expressed, is to allow for the creation of digital personas from historical collections, old videos or even family photo albums—using images to create an interactive model of someone who is far away.
“With the ability to digitize anyone,” says Supasorn Suwajanakorn, a graduate student at UW and lead author of the paper to be presented in Chile in December, “you can replay the memory of your beloved ones or interact with them as if they were with you.”
However, a major component of the breakthrough is also the ability to use older materials.