From a very early age we learn to recognize faces
and the significance of particular expressions. The signals take just a few
milliseconds to read. Our dogs are very good at this too. They read us
sufficiently accurately to convince us that “they understand every word I say”
even when we really know that we’re not speaking dog.
Of course, we say (maybe), that such recognition
skills are critical for survival and that we’d have never made it this far
without them, but at the same time we can also scare ourselves on dark nights
in the woods by seeing faces in vague, moving bushes. Seeing faces in the
clouds moving across the sky or in the patterns on tiles on the bathroom floor
are much more relaxing, of course. The question arises, though, what are the
critical features of the pattern that causes us to think, “face”?
Immediately after that thought, comes the one about
how to train one’s computerized camera to recognize faces. Currently the
fastest algorithm for electronic face recognition is the ‘Viola-Jones’
algorithm developed by Paul Viola and Michael Jones ten years ago. The accuracy
of face recognition is now good enough for the process to be found in digital
cameras.
A current paper by Hart et al from Phillips U at Marburg explores the similarity for
patterns, which are not faces, being seen as faces by both people and machines
(1). The humans were two-dozen young students who were permitted a glance at a
series of blurry pictures. Some were of faces and some weren’t. There was only
one machine in the experiment. Machines, of course, don’t come with individual
preferences, only algorithms (don’t they?)
It turned out that people and the machine were
quite good at picking a real face when paired with an illusory face. Not
perfect, but no surprises there.
If the images are partially obstructed by vertical
bars, then the mistakes increase significantly. Could be good pointers here for
camouflage experts and those putting on war paint. Of course, this wasn’t the
discussion point of the paper, rather the authors suggest that our brains use a
similar analysis tree in deciding on a face being a face.
So next time you’re scared by something lurking in
the bushes on that dark stormy night, take a picture and see if your camera is
correcting for ‘red-eye’. If it is, run! It’s always good to have a second
opinion.
