Sci TechBits reports on how face synthesis software could become the next big step in mainstream avatar association. With two cameras focused on the edge of the person’s TV or computer screen, capture and analyze the data that the cameras capture, and try best to fill in any missing gaps, then synthesize that date and match it to a library of pre-existing facial expressions that would best compliment the character’s reaction on screen to the person sitting in front of the cameras.

There are a few roadblocks this technology has that the site acknowledges on. One is that with this technology, usually you can’t capture the emotions of the player as quick as it would like, simply because of how fast the person might change their stance during gameplay, which is why the latest revision in this genre of facial recognition technology relies on having the person input specific facial commands ahead of time before putting the technology to work so that it has the previously mentioned library of expressions to recognize and employ.

Second, as noted in the comments of the story is that any sort of generated facial avatar would still look pretty damn cartoony and uncanny. Technology is still trying to find a way to eliminate that uncanny valley look that the faces generated would mostly be finding themselves in. One such solution put forth by Daniel Floyd is to stay on the stylization side of the uncanny valley for all it’s worth and the benefits that not taking the jump into photo-realism entails. Because seriously when I first looked at the photos of sample faces on the site I was like “GAH!” Either that one on the far right is trying to be stylized and succeeding, or trying to be photo-realistic and ending up looking like the love child of a Shrek Character and a wrestler. – The Ben