Thursday, April 30, 2009
A camera that knows who it's photographing
I know a little bit about complex pattern recognition, and it's not easy or straightforward. So my jaw dropped a bit when I read David Pogue's footnotes on his review of the Panasonic Lumix GH-1 camera. What got me was this:
Coming soon, self-tagging photos, a speech-to-text dictation machine, caller ID by voice, nav-by-video cars, et cetera. Sheesh, the future is more evenly distributed every day.
The Lumix doesn't just recognize that something *is* a face; it actually tries to determine *whose* face it is. That way, it can give priority of focus and exposure to your own friends and family, even plucking them out of a crowd.
What the marketing materials don't say is that you first have to register each person you want it to recognize. The camera displays a little on-screen template, basically a box with two eyeholes; you're supposed to take a snapshot of the person, facing front and positioned to match the template. You can even name the person, using an onscreen keyboard; if you register more than one person, you can rank them by priority. (Dealing with the hurt feelings if someone catches you ranking them low is left to you.)
Coming soon, self-tagging photos, a speech-to-text dictation machine, caller ID by voice, nav-by-video cars, et cetera. Sheesh, the future is more evenly distributed every day.
Technorati Tags: DavidPogue, Photography
Comments:
David,
To extend your bemusement---take a look at the most recent iteration of iPhoto. It's facial recognition actually works. As someone who once obsessively read the literature on all sorts of machine recognition I am stunned by how well. Now it's operating in a limited domain and that's part of how it is able to achieve a usable level of "right" calls but then we too only operate in a limited domain ourselves. (smile)
Google has an image recognition engine out recently too...it too is limited and in addition relies on textural clues to focus its hits but the same caveat about ourselves holds true.
The interesting moment will arrive when the producers of such apparatus move on to using our preferences and contexts as exemplified in our online actions to shape the returns...(which is most of how iPhoto actually works since it is limited to a single user.)
A new spin on privacy issues then emerges....
Cheers!
Post a Comment
To extend your bemusement---take a look at the most recent iteration of iPhoto. It's facial recognition actually works. As someone who once obsessively read the literature on all sorts of machine recognition I am stunned by how well. Now it's operating in a limited domain and that's part of how it is able to achieve a usable level of "right" calls but then we too only operate in a limited domain ourselves. (smile)
Google has an image recognition engine out recently too...it too is limited and in addition relies on textural clues to focus its hits but the same caveat about ourselves holds true.
The interesting moment will arrive when the producers of such apparatus move on to using our preferences and contexts as exemplified in our online actions to shape the returns...(which is most of how iPhoto actually works since it is limited to a single user.)
A new spin on privacy issues then emerges....
Cheers!
