I have been thinking a bit more about Picasa’s face recognition. Steve observed that his experiments with unnamed files from his camera suggest that it does not seem to rely on file names, so we are back to the explanation of magic.
However, like most magic, I suspect there is a good dose of sleight of hand involved. I was tremendously impressed by the amount of pictures Picasa identified with better than 99% accuracy (two misplaced images in over two hundred that it picked out). In fact, I was so impressed that I launched straight into wading through the remaining faces and adding further names or associating them with existing people and have now whittled the remaining pool down to about three hundred from nearly two thousand.
That means I now have a much better index of where to find people in my photo collection than ever before but I’ve done a lot of the work myself. Also, while Picasa has been remarkably accurate at picking out faces for consideration (very few thumbnails showed things that weren’t people’s faces and the majority of those were faces in paintings or sculptures), I can’t easily put my finger on how many pictures with faces it hasn’t included.
So, with me doing a lot of the work and no way to easily check if I am really down to the last batch of photos with faces, some sleight of hand is involved. However, it could also be argued that the software design has inspired me to press on with the task and made it painless or even enjoyable so still a kind of magic after all.