Good with faces
Aug. 18th, 2006 10:18 am![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
imomusReading newspaper reports of crimes, I'm constantly amazed at the ability of witnesses to recall the faces and clothes of suspects, make police robo-portraits, and pick out culprits in identity parades. Of course, sometimes DNA testing or circumstantial evidence later reveals this kind of certainty as misplaced. But I must say it's rather enviable.
I wish I were better with faces. It's a totally normal thing for me to be approached by a stranger saying "Hello, we met at such-and-such, perhaps you don't remember me..." And they're absolutely right, I don't. I think one reason for this is that I'm rather introverted and don't make much eye contact. So even though I might be with someone for ten minutes, that might only include a few seconds of looking directly at them, and those few seconds are difficult to access a year later. I certainly don't make conscious mental notes like "Has alert, greeny-blue eyes, sparse eyebrows, brown hair, wide-spread lips finely-pointed at the ends, square jaw..." I'm also rather in despair of novels which provide such descriptions, because putting together a face in your mind from verbal descriptions is one of the most futile activities known to man, and a great demonstration of the limits of language.
But over the last couple of days I've got rather more confident in my ability to "recall" faces thanks to a Java applet called FaceMachine. Developed by Tom Busey at Indiana University, FaceMachine uses "Eigenfaces", collections of features that tend to occur together in faces. New faces are created by adding and subtracting Eigenfaces from the Average Face. (You can download a version which allows you to save the faces you create here.)
The fascinating thing -- as you slide FaceMachine's 15 sliders back and forth, trying to get the face to look like someone specific -- is that you aren't using observation, or working on specific facial details as you would when drawing someone from life. It's much more intuitive. I found myself simply asking "Does this look more or less like X?" as I moved the sliders around. Tiny incremental decisions like that, not based even on a specific mental image of what X does look like, ended up yielding some very accurate representations. See if you can guess who the images I came up with are supposed to be.
Some faces were more difficult than others. My own, for instance, was tricky. David Bowie's was suspiciously easy (he kept popping up when I tried to make Tony Blair too), which suggests either that his face is some kind of archetype, or that it's been fed into the software as one of the sample faces. FaceMachine generates women and East Asian faces too, but these seemed more difficult to control successfully.
I'm interested in the Gladwellian aspects of all this; the idea that we can think rather effectively using the unconscious and intuitive parts of our brain, rather than the empirical, observational parts. In "Blink: the power of thinking without thinking", Malcolm Gladwell looks at the two-second snap judgements we all make. We make them without sufficient "evidence". We make them in traffic, and our lives depend on them. Sometimes they're wrong, and they kill us. But a lot of the time they're right, and we survive to snap judge another day.
Gladwell actually doesn't like the word intuition; for him, a snap judgement is just rapid, rational thinking, sometimes right, sometimes wrong. I found myself making a succession of snap judgements as I slid the FaceMachine sliders. Those micro-judgements (does that look more like the person, or less?) accumulated into portraits as accurate as I could have come up with by looking at someone's face and drawing it, holding my pencil up to measure dimensions in units.
Another way of saying this: I may not know what you look like, but I can recognize when I see it... and when I don't. And a series of those recognitions can lead to a surprisingly acute representation of your face. As a musician, I work very much the same way. I can't specify in advance what I want something to sound like, but when I start listening to sounds I can very quickly reject the "wrong" ones and select the "right" ones. If you asked me to justify each decision I'd have a hard time explaining why those things are wrong or right, but my quickfire certainty is a big part of the way I operate as an artist. Nope, nope, nope, YES. Why? No idea.
I wish I were better with faces. It's a totally normal thing for me to be approached by a stranger saying "Hello, we met at such-and-such, perhaps you don't remember me..." And they're absolutely right, I don't. I think one reason for this is that I'm rather introverted and don't make much eye contact. So even though I might be with someone for ten minutes, that might only include a few seconds of looking directly at them, and those few seconds are difficult to access a year later. I certainly don't make conscious mental notes like "Has alert, greeny-blue eyes, sparse eyebrows, brown hair, wide-spread lips finely-pointed at the ends, square jaw..." I'm also rather in despair of novels which provide such descriptions, because putting together a face in your mind from verbal descriptions is one of the most futile activities known to man, and a great demonstration of the limits of language.
But over the last couple of days I've got rather more confident in my ability to "recall" faces thanks to a Java applet called FaceMachine. Developed by Tom Busey at Indiana University, FaceMachine uses "Eigenfaces", collections of features that tend to occur together in faces. New faces are created by adding and subtracting Eigenfaces from the Average Face. (You can download a version which allows you to save the faces you create here.)
The fascinating thing -- as you slide FaceMachine's 15 sliders back and forth, trying to get the face to look like someone specific -- is that you aren't using observation, or working on specific facial details as you would when drawing someone from life. It's much more intuitive. I found myself simply asking "Does this look more or less like X?" as I moved the sliders around. Tiny incremental decisions like that, not based even on a specific mental image of what X does look like, ended up yielding some very accurate representations. See if you can guess who the images I came up with are supposed to be.
Some faces were more difficult than others. My own, for instance, was tricky. David Bowie's was suspiciously easy (he kept popping up when I tried to make Tony Blair too), which suggests either that his face is some kind of archetype, or that it's been fed into the software as one of the sample faces. FaceMachine generates women and East Asian faces too, but these seemed more difficult to control successfully.
I'm interested in the Gladwellian aspects of all this; the idea that we can think rather effectively using the unconscious and intuitive parts of our brain, rather than the empirical, observational parts. In "Blink: the power of thinking without thinking", Malcolm Gladwell looks at the two-second snap judgements we all make. We make them without sufficient "evidence". We make them in traffic, and our lives depend on them. Sometimes they're wrong, and they kill us. But a lot of the time they're right, and we survive to snap judge another day.
Gladwell actually doesn't like the word intuition; for him, a snap judgement is just rapid, rational thinking, sometimes right, sometimes wrong. I found myself making a succession of snap judgements as I slid the FaceMachine sliders. Those micro-judgements (does that look more like the person, or less?) accumulated into portraits as accurate as I could have come up with by looking at someone's face and drawing it, holding my pencil up to measure dimensions in units.
Another way of saying this: I may not know what you look like, but I can recognize when I see it... and when I don't. And a series of those recognitions can lead to a surprisingly acute representation of your face. As a musician, I work very much the same way. I can't specify in advance what I want something to sound like, but when I start listening to sounds I can very quickly reject the "wrong" ones and select the "right" ones. If you asked me to justify each decision I'd have a hard time explaining why those things are wrong or right, but my quickfire certainty is a big part of the way I operate as an artist. Nope, nope, nope, YES. Why? No idea.
![[identity profile]](https://www.dreamwidth.org/img/silk/identity/openid.png){kind=small}
(no subject)
Date: 2006-08-18 02:21 pm (UTC)(no subject)
Date: 2006-08-18 02:41 pm (UTC)(no subject)
Date: 2006-08-18 02:42 pm (UTC)(no subject)
Date: 2006-08-18 06:24 pm (UTC)I dislike him, but for other reasons....
Date: 2006-08-18 04:44 pm (UTC)http://gladwell.typepad.com/gladwellcom/2006/04/the_duke_case.html
Re: I dislike him, but for other reasons....
Date: 2006-08-18 07:30 pm (UTC)Eyewitnesses are patently unreliable, especially after incidents of severe emotional trauma.
Re: I dislike him, but for other reasons....
Date: 2006-08-18 07:48 pm (UTC)spineless?? u gotta be kidding
Date: 2006-08-19 03:48 am (UTC)I love Gladwell's pieces in the New Yorker. Though I can't say I ever fully understood the tipping point. Reaching a tipping point: good? bad? Tough to say. Too conceptual for me, I think.