That painting above, of the Madonna and child? Art experts have always suspected that Italian Renaissance master Pietro Perugino didn't paint it entirely by himself. But they were never entirely sure.
Now a team of Dartmouth professors say they've mathematically proved that several different people worked on the painting. They developed a technique in which they digitize the painting into a huge 16,852-by-18,204-pixel photo. Then they took the faces of the six people in the painting, and broke them into several hundred sections, 256 by 256 pixels in size. Then, as Wired News reports:
Art scholars are dubious that the technique is useful, but personally, I'm intrigued by it. It seems like an interesting literalization of the brain's statistical data-processing equipment. We humans are incredibly good at knowing when faces just don't quite match up or don't seem quite right. Though the scientists have picked a bunch of obviously arbitrary mathematical markers, the idea of data that doesn't "match up" seems like a neat metaphor or analogue for what the brains of art experts are doing when they look at the paintings: They're crunching the patterns, comparing them to the enormous database of all art they've seen before, and detecting something -- oh so subtle -- amiss.
Posted by Clive Thompson at November 25, 2004 02:07 AM
Trackback Pings
TrackBack URL for this entry: http://www.collisiondetection.net/mt3/mt-tb.cgi/1039
Carefull...
The point is WHAT algorithm?
I mean what is the RULE they aplly that tells the amount of difference between two different faces?
It's easy to say "this algorithm tells the difference" but even if it gets it right a given number of times it doesen't mean that the underlying theory is right. If there is an underlying theory...
What I mean is that too many times researchers (specially in this field) are hiding behind the mantra > that means nothing.
...In the loving memory of Karl Popper...
PS
I confess I didn't read the linked article. I wil, promised... ;D
Posted by: Mario at November 25, 2004 2:43 AM
Well, like I said, I think the data points could be almost arbitrarily picked, so long as they produce consistent differences across the painting. At that point, I'd imagine many different off-the-shelf compare-and-contrast or clustering algorightms would work fine. But yeah, they don't clarify particularly what approach they use in the article!
Posted by: Clive at November 25, 2004 3:22 AM
I like your analogy regarding human face recognition. It's interesting to consider what 'algorithms' our brains use to detect conspicuous anomalies. And although the scientists on this project picked a set of arbitrary markers, isn't it possible that our brains do something very similar. There is no reason to assume that salient differences are biologically relevant; only the outcome of noticing the differences is relevant.
Some researchers have argued something along the lines that all you need in order to recognize a face are the eyebrows. Seems arbitrary to me.
Regardless, I'm unconvinced that this software does what it claims to do. If differences exist in the pixelation between to faces isn't it possible that it's the product of intention, or accident, by a single artist, rather than a group of different artists.
Maybe it does work though. Humans and computers are often good at processing different information. Humans are great at recognizing faces, and horrible at detecting artistic forgery I'm guessing. Computers suck at recognizing faces; maybe art forgery is their milieu.
Posted by: Steve at November 25, 2004 7:22 PM
Is the Madonna wearing blue jeans?
Posted by: jim at November 28, 2004 6:56 PM
The BLU JEANS!
That's how it detects forgery!!!
:D
Posted by: Mario at November 29, 2004 3:40 AM
Heh.
Steve, yep, I think that's precisely right -- computers are incredibly bad at recognizing objects, while humans excel at it. But humans suck at doing finely-adjustable analysis of the statistical vectors of a picture -- i.e. we're good at recognizing statistical patterns, but only the ones we're trained or hard-wired to see; we cannot easily adjust our parameters to see new patterns in the data. But computers are great at that.
Posted by: Clive at November 29, 2004 6:32 AM
could it be at all possible that, oh i dont know, maybe the same guy painted everything but he wanted more points on certain things, perhaps because he wanted them to be the focus?
Posted by: woot at December 1, 2004 1:43 PM
Posted by: online poker at January 27, 2005 4:43 AM
Posted by: online poker at January 28, 2005 10:10 PM
Posted by: online poker at February 1, 2005 3:17 AM
Posted by: online poker at February 1, 2005 9:30 PM
Posted by: payday loan at February 9, 2005 12:54 AM
Posted by: online poker at February 10, 2005 3:18 PM
Carefull...
The point is WHAT algorithm?
I mean what is the RULE they aplly that tells the amount of difference between two different faces?
It's easy to say "this algorithm tells the difference" but even if it gets it right a given number of times it doesen't mean that the underlying theory is right. If there is an underlying theory...
What I mean is that too many times researchers (specially in this field) are hiding behind the mantra > that means nothing.
...In the loving memory of Karl Popper...
PS
I confess I didn't read the linked article. I wil, promised... ;D
Posted by: Mario at November 25, 2004 2:43 AM
Well, like I said, I think the data points could be almost arbitrarily picked, so long as they produce consistent differences across the painting. At that point, I'd imagine many different off-the-shelf compare-and-contrast or clustering algorightms would work fine. But yeah, they don't clarify particularly what approach they use in the article!
Posted by: Clive at November 25, 2004 3:22 AM
I like your analogy regarding human face recognition. It's interesting to consider what 'algorithms' our brains use to detect conspicuous anomalies. And although the scientists on this project picked a set of arbitrary markers, isn't it possible that our brains do something very similar. There is no reason to assume that salient differences are biologically relevant; only the outcome of noticing the differences is relevant.
Some researchers have argued something along the lines that all you need in order to recognize a face are the eyebrows. Seems arbitrary to me.
Regardless, I'm unconvinced that this software does what it claims to do. If differences exist in the pixelation between to faces isn't it possible that it's the product of intention, or accident, by a single artist, rather than a group of different artists.
Maybe it does work though. Humans and computers are often good at processing different information. Humans are great at recognizing faces, and horrible at detecting artistic forgery I'm guessing. Computers suck at recognizing faces; maybe art forgery is their milieu.
Posted by: Steve at November 25, 2004 7:22 PM
Is the Madonna wearing blue jeans?
Posted by: jim at November 28, 2004 6:56 PM
The BLU JEANS!
That's how it detects forgery!!!
:D
Posted by: Mario at November 29, 2004 3:40 AM
Heh.
Steve, yep, I think that's precisely right -- computers are incredibly bad at recognizing objects, while humans excel at it. But humans suck at doing finely-adjustable analysis of the statistical vectors of a picture -- i.e. we're good at recognizing statistical patterns, but only the ones we're trained or hard-wired to see; we cannot easily adjust our parameters to see new patterns in the data. But computers are great at that.
Posted by: Clive at November 29, 2004 6:32 AM
could it be at all possible that, oh i dont know, maybe the same guy painted everything but he wanted more points on certain things, perhaps because he wanted them to be the focus?
Posted by: woot at December 1, 2004 1:43 PM
7695 http://www.online-poker-web.net
online poker
Posted by: online poker at January 27, 2005 4:43 AM
4755 want to play online poker mate?
Posted by: online poker at January 28, 2005 10:10 PM
4542 want to play online poker mate?
Posted by: online poker at February 1, 2005 3:17 AM
1766 want to play online poker mate?
Posted by: online poker at February 1, 2005 9:30 PM
1698 http://www.online-poker-e.com
online poker
Posted by: payday loan at February 9, 2005 12:54 AM
2754 http://www.online-poker-now.com
online poker
Posted by: online poker at February 10, 2005 3:18 PM