I found an article at the Aperture Users Network which discusses and explains the Faces 'tool' in iPhoto09 and thought I'd share it here. As its author, David Schloss, says, the concept builds on the way the human brain allows us [or, at least some of us, since I'm not often right] to identify people over time. I thought this was particularly fascinating:
The human brain is specifically designed to find faces, it's an evolutionary trait that's required to help us differentiate between our parents and a hungry lion. It also happens to be a bit of a tricky task for a computer to manage thanks to the changing nature of the human face.
Generally speaking facial recognition works by figuring out the geometric patterns between key elements in the face and assigning values to those patterns. Anything that has the same pattern is probably the same person.
But think about the huge variability of people over time. Children age, changing their faces both lengthwise and in width. As we age our features sag or get thinner or thicker–the geometric patterns between our nose and our chins might not change, but anyone who has seen a show like "The Biggest Looser" knows how radically a face can change over time. And most photo albums have photographs of people over a vast speriod (my wife, for example, is in photographs going back until January 2000, when I put my first image in iPhoto.
But not only are faces able to morph over time, they're often distorted by the lens, covered by objects like glasses headwear, scarves and many other objects that our brain just works around. Multiply that by the various directions the face is viewable from (a profile looks different than a straight-on head shot) and you see the enormous challenges inherent in finding a face in a crowd.