I first need to draw a distinction between Facial Detection and Facial Recognition. Facial Detection is common on still and video cameras to detect the presence (or not) of one or more faces. Facial Recognition provides a name for each of the faces.
Still cameras use Facial Detection metadata to focus on the face, making the assumption that the face will be the subject, unless told otherwise. Both Final Cut Pro X and Premiere Pro CC use Facial Detection.
Facial Detection is used in Final Cut Pro X to detect how many people are in the shot and the size of the faces that it detects. (Final Cut Pro X then Infers the type of shot — W, M. CU — by the relative size of the detected face in the shot, but that’s for the next chapter.)
Facial Detection is useful metadata as Final Cut Pro X demonstrates, but Facial Recognition – where the software attempts to identify the individual whose face is in the shot – is more valuable, if accurate.
The current state of the art is that Facial Detection is very advanced, but Facial Recognition still has a way to go. It seems to be most accurate identifying many shots of the same person in a limited period of time, but less accurate at detecting the same person when there are many examples over an extended period of time. The aging process seems to confuse the current technology.
In September 2010 Apple purchased Swedish Facial Recognition company, Polar Rose to further enhance its Facial Recognition technology base, but we haven’t really seen any improvements in iPhoto or Aperture since then, Perhaps the upcoming Photos app will have improved Facial Recognition.
Facial Recognition and Facial Detection are both useful forms of metadata but having the person identified is much more valuable, if it’s accurate. In the age of social media there are ways we can find out names if we have accurate Facial Recognition. See Outside Sources for algorithmically derived Content Metadata.
Facial Detection and Recognition have both been limited to front-on views, but new technology has been developed to at least detect a face, even if side on.
Return to The Future of Content Metadata….