Some days ago I've received an e-mail from a dear friend, in which there was a photo collage of her, in the middle, and a bunch of celebrities whirling around her photo.
Each photo presents the celebrity name and a "look-alike percentage".
You can see a live example in the sidebar of this blog, just scroll down a bit.
I found it really funny and entertaining, and I started feeding the matching algorithm with a lot of pictures of myself (tanned, recovering from illness, with glasses, a slim period, a fat period, drunk, really drunk...)!
It is not obvious, but each photo carries certain features that make a different match set, therefore you are not always Mel Gibson's secret brother :).
Anyway, it is interesting to note that several celebrity matches are always there, in the top ten, and that tells me that a GOOD face recognition system MUST work with a photo set of the candidates, not only a picture. (I've been working on classifying algorithms for a while, and I personally experimented the importance of considering the variances along with the "average" model, i.e. don't blindly trust the first match result)
Another interesting thing that smells "viral marketing" is the fact that this "nice free face recognition tool" carries a promotional message, which says: Myheritage.com is building a social network of families, and is competent on face recognition products.
Instead of building yet-another-youtube-promo-video, they built an application. A nice one, that users will remember to use and spread. As a consequence of that, I imagine they are seeing a great increase in their subscription daily rate...
Technorati Tags: cernio, myheritage, face recognition, marketing20
