Friends, Facebook has just taken facial recognition to a whole new level.
Through its overly-creepy-sounding project "DeepFace," Facebook has been exploring ways in which to improve its facial recognition software -- all the way to the point of "human-level performance." How? The new software maps a person's facial features in 3-D, then turns them into a flat model. The results are then filtered by color to highlight specific facial elements, such as nose or brow shape.
According to Facebook, their approach is 97.25 percent accurate, which is very close to human accuracy levels. In fact, there are only a few factors keeping software from exceeding human facial verification accuracy, including lighting, expression, pose and image quality. However, with enough time perhaps Facebook's DeepFace team can overcome even those.
For now, Facebook users won't be impacted by the new software. First, it's heading to the IEEE Conference on Computer Vision and Pattern Recognition in June to be vetted by other researchers, then it will probably have to face the music with privacy advocates. How exactly will Facebook then use the software? We're not entirely sure, but it's probably safe to assume that facial recognition this powerful probably won't be used only for tagging people in uploaded photos.