Facing Facts

When is a face not a face?

One of the benefits of living in the House of Googleeven if it's only in a derivative, I'm-using-their-software kind of wayis the assortment of nifty tools and applications they make available, all for the price of a six-pack of nothing. Although I haven't yet investigated their new browser or the Picasa 3 photo organizing and editing software, I spent some time over the weekend exploring the world of Picasa Web Albums. In particular, its image-recognition capabilities suggest new and intriguing possibilities when you consider how many digitized faces are stored on the servers that make the Web go 'round.

Zillions of faces, I think. But who are they? Where are the labelsthe proper namesthat allow us to make the distinction between Robert and Larry, or Katherine and Isabella? The answer, of course, is I don't know. Who has the time or inclination to catalog every single face in every digital snapshot that, these days, invariably winds up on the Internet?

I may know the answer to that one, actually. Google has the time, and the resources, to make this sort of thing feasible. Using Picasa Web Albums as the human/machine interface, and assuming we can free up enough timeevery so oftento type in the names of our photographic subjects before we send their images to the Web, Google's software will keep track of those faces and their associated names, thereby freeing us up for more photography.

At least in theory. The four images you see at the top of this monologue are actual examples of graphicsfrom the Omegaword blog in this caseselected by the Picasa Web Albums system as possible faces for me to identify and name. As you can see, the first two are essentially line drawings, from which Picasa's image-recognition system was evidently able to extract sufficient humanity to make the call. The third image, on the other hand, is possessed of only rudimentary facelike characteristics, making the system's judgment all the more impressive. As for the rightmost image, well . . . the system isn't perfect. I can see it's a button, and you can see it's a button, but then, we aren't machines. At least you're not.

One especially disturbing moment occurred when the image-recognition system suggested I might be the subject of a particular portrait I had just uploaded, when in fact it was a picture of my mother. But on the whole, using actual humans in actual photographs worked surprisingly well. Once I had correctly identified a given subject in a photograph, the Picasa system did a good job of guessing his or her name in subsequent photos.

At some pointand on a much grander scaleselecting return images that contain faces in Google's advanced image search, after typing in a particular name, may become the method of choice for keeping track of anyone who's ever been photographed. Where exactly that point might lie on our timeline is uncertain, but considering the sheer number of wired and wireless cameras on our planet, it could be nearer than it seems today.

I also can't resist mentioning, in passing, that the procedure in the previous paragraph didn't work out well for me. Using Omegaword as the search phrase, Google's face-search returned mug shots of other people, while ignoring my mug entirely. I'll leave the reenactment to you if you're so inclined, and so bored.


No comments:

Post a Comment