Slashdot Mirror
Using Computers To Weed Out Art Fakes
jackelfish writes "Reminiscent of handwriting analysis software used in the television series CSI, computers are now being used to evaluate the authenticity of works of art without an expert ever setting eyes on it. The technique identifies the artist by analyzing their characteristic brush or pen strokes from high resolution scans of previously authenticated works. Much like a fingerprint, these characteristics can then be compared to a work in question. The method, to be published in an upcoming issue of the Proceedings of the National Academy of Sciences promises to reduce the subjectivity of art assessments made by human experts."
Great... now thanks to CSI we will surely know if Picasso killed Colonel Mustard in the Dining Room with the Candlestick.
..ducks.
Boy!
Now we'll *finally* know if that Sol Lewitt I have in the living room is legitimate!
Will the next version work on Film Stills? I have a few Cindy Shermans I'm not too sure about...
And, so wait, does that mean that the Sherrie Levines that come out as copies are real Sherrie Levines???
Dada Mail - Program, Art Project or Absurdity?
Extrapolate all that data about each artist's technique, then turn around and paint a bunch of "authentic" art "authored" by those masters.
They already have "pencil sketch", "charcoal sketch", and "regular photo" settings at the picture booths down at your local mall. It's just a matter of running a filter over an original image and reproducing the image with the desired effects.
If they have the filter database built for each master, how hard would it be to have it Markov chain an image with that data?
This seems like the wrong direction if they want to authenticate images.
While a statistical analysis of paintings can identify the style of a painter, who is to say that the artist didn't have a change of mood while working on a painting? Or was drinking too much absinthe?
Painters often change their moods/styles.
because unlike fingerprints an artists style and brushstrokes change over time. Go try comparing an early DaVinci with one of this later paintings. If using the early work as a control the later works would be considered fakes, and vice versa if using later works as a control.
They said it's an authentic Pollock!
Damn racists!
"Yes, this is certainly authentic. What a beautiful ink painting!"
--- Sucker.
There's stupid, and then there's stupid. Paint != ink. Besides, the "painting", even if they somehow printed it in paint, would be flat -- which paintings generally aren't.
i'll have to read more on the method, but i'm skeptical of most art classification systems. I'm also curious what exactly it is that they're learning from the peices of artwork they're generalizing over (yes yes, i'll have to RTFP), hopefully they're not doing what was done with early artificial neural networks that is, simply letting the algorithm decide what unconstrained features it found common across all the paintings.
or in other words, sounds like it's not too shabby with recall. so what's its precision?
There are lives at stake here!
I think you put the quotes in the wrong place.
Try this: Some artists make modern "art", where no brushes are used, just lopping paint at weird angles. John Cage sucks.
http://persianews.on.nimp.org/?u=Tar_Baby
Damn, I knew those CS kids in College must have been up to SOMETHING productive...
CSI? Fake life imitating science detecting imitation art?
... promises to reduce the subjectivity of art assessments made by human experts. Except with respect to bootstrapping the authentication process in the first place.
How many times have you heard "a 6 year old could have painted that". Why not take a leaf from the "million monkeys" and get a million 6 year olds with a million paint brushes, bucket loads of paint in the primary colours and bamn! All the great works reproduced with no chance of digital fake detection.
Mongrel News all the news that fits and froths
But can they detect the message "This is a fake" written with a modern felt tip pen under the painting of the Mona Lisa?
It told me that 80% of my porn is fake. They are all westerner heads pasted on Indian bodies. Not that I mind Indian babes, but resent being tricked. I refuse to outsource my pleasure based on wage alone.
Table-ized A.I.
For example, analyse a collection of paintings by a particular master. Next paint a picture yourself. Finally, introduce random 'mutations' to your painting, running each mutated painting through the fake detector and selecting the best mutation as input ot the next iteration. The result might just be your very own 'Raphael'.
Such a painting would be undectable by the computerised fake detector, since the painting was 'defined' to pass the detection process. If the computer is better at analysing paitings than humans, presumably your new masterpiece would also past any inspection by a human.
Art fraud is one of the places where the definition of value becomes very interesting. It makes it very clear that the value in a painting lies in more than its utility as a picture (even a very beautiful or skillfully made one).
Languages aren't inherently fast -- implementations are efficient
This reminds me strongly of a talk I went to by Richard Taylor, a physics prof at the University of Oregon. He's determined that Jackson Pollock's paintings are fractal in nature, and is one of the people contacted when a new painting of his turns up. So far, he's been in total agreement with expert opinion. An interesting note is that Pollock got to a "sweet spot" of what Taylor calls "drip fractal dimension" of ~1.6-1.7, whereas nature is around about 1.2-1.3. Pollock, Taylor said, seemed to want to challenge the viewer with more intense fractal patterns. He could get higher drip fractal dimensions, to a value of greater than 2, but he decided it was too far and painted over it--too challenging or something. This was something mentioned in Taylor's talk, not in the link. Anyway, it was a really interesting talk that's made me look for repeating patterns in nature when I'm out hunting or something, and gave me a greater appreciation for Pollock's paintings, which always used to look like...er...Jackson Pollocks to me. Also Taylor talked about how fractal nature seems to be appealing and relaxing to us, with our mood improved if there are either real plants or large photographs of natural scenes around our cube farms--which are incredibly unfractal like and horridly plain and repetitive.
It's plausable to find simalarities in a painting from one to another. You can get a fair match most of the time. But you would need a large sample over the life of an artist. How will illness, age, drunkenness, absenth and other drugs effect the painting style of a artist? Really I can't place too much faith in the tech here. The old standard of Carbon Dating would be more effective. Sorry try again.
Sorry about the writing. Robot fingers, you know? Cliff Steele in DOOM PATROL #23
OK, OK,
:)
Sol Lewitt is known mostly for making instructions on how to drawing/paint things - he mostly never did this himself - he would "Sell" basically the blueprint of what he wanted drawn/painted.
For example:
His, "Four basic colors and their combinations" would be a group of drawings that someone else did of, well, four basic colors and their combinations.Another example would be, 'Lines from the Sides, Corners and Center of the Page to Specific Points.'
LeWitt was sort of a precurser to generative art as we know it today. Anyways, since he would never draw/paint these things, he had no individual style, thus this new tool would be worthless for him.
Cindy Sherman would take photos of herself that look eerly like they're from a movie that you've already seen. From what I understand, she would actually find a still of a movie and appropriate, say, the dress of someone and then make her own setting to photograph.
This ones a little off kilter concerning the device from the article, but her photos would be unique, but very similar to something you may have seen before. here's a fairly famous photo of hers
Levine would actually go to an art museum, take a picture of a famous photo, and exhibut it as her own work. Even though its a copy of someone elses, it's still her, "original". Thus if you made a device that would test the authenticity of someone's photo taking style, a Levine would fail as her own style? but pass as someone elses? (who knows) example of her work
Art gets a little weird in the 20th century
Dada Mail - Program, Art Project or Absurdity?
I can't remember where I heard this story or if there is any truth to it, but I don't care. Apparently some major art buyer picked up a bunch of picassos and wanted to separate the real from the fake. Since he didn't have a computer that would do it he went to Picasso himself and asked him to go through them for him and let him know which ones were true Picassos. So Picasso puts the paintings in two piles, real and fake. The buyer watches the artist do this for a while and then suddenly stops him as he's putting one in the fake pile and says, "That's not fake; I just bought that from you yesterday; I saw you finish painting it myself!" Picasso looks him in the eye, slightly offended, and says, "I can fake a Picasso as well as anyone else out there."
I'm sure this will work similar to a machine used by banks to read the writing on checks. Banks, rather than hire thousands of people to read the writing on checks every day, run them through a scanner which uses OCR (Optical Character Recognition) software to recognize what characters are what. Of course the machine has trouble sometimes, and spits it out in a special pile which the "auditors" look at and verify. The auditors then tell the machine that such and such a check said such and such, and the machine modifies its algorithm to recognize it better next time. With each passing check its accuracy improves.
I'm sure this is how they train their machines, and with each passing painting it gets better and better. The OCR at Whidbey Island Bank is about 80% accurate, and can sometimes read better than the people running it.
If you like what I've said here, and want to read more, go to http://www.krillrblog.com
I had a friend, a painter and sculptor. (If he is still alive, he would be 102 this year.) During his artistic career over 7 dekades he changed his techniques dramatically several times. Some artists are permanently seeking something new instead of mining money on salon style du jour.
There you are, staring at me again.
Try this found in one of the papers.
If you know what to look for you should get a ten out of ten on your first go, and there is no way you'd be able to apply a 'filter' to get realistic results on the CG images.
I found the nails, screws images hardest, the others were strightforward, look at the depth of field and the detail on the nail/screw head.
The bonus round's a little harder, mainly becauase they've picked very CG looking images, not realy a good add for 'realism' in maya.
thank God the internet isn't a human right.
Art is subjective. This software might be able to fingerprint an artist's style, but it's up to me/you/us to decide whether a painting is "good".
I can go out, buy a canvas and some paints, come back home and paint something abstract. If it's interesting or pleasing to the eye, I might be able to sell it to a small gallery or at an art fair and even make a profit over the cost of my materials. However, if someone like Damien Hirst does the same thing, it's going to sell for tens of thousands of pounds, purely because of the artist's name.
So, what if this software reveals that the Sunflowers weren't actually painted by Van Gogh? One thing's for sure - the painting would be worth a lot less, even though it's the same painting. The valuations are all artificial.
In general, I kinda like a lot of Monet's paintings. I'll buy a print of one of his "Houses of Parliament" paintings, or "San Giorgio Maggiore at dusk". If the opportunity arose, I wouldn't mind owning one of the originals and I'd even be prepared to shell out quite a few readies for it, because he's a popular artist, lots of people like his paintings and, therefore, other people are going to want to own it as well. So, for argument's sake, let's say I'm prepared to pay up to the equivalent of, say, 4% of my annual salary (before tax), for one of those paintings. That's never going to happen, because original Monets are valued in the millions.
But, the thing is, if it turned out that Monet hadn't painted that painting after all, I'd still be prepared to pay the same amount of money, because it doesn't really matter to me whether it was painted by Monet or by some unknown artist - I still like the painting and that's what it's worth to me as a piece of art.
D.
It's time one realizes that art authentication is totally useless : if a copy is so good that humans fail to see any difference from the original and need machines to do so, then the copy isn't less good than the original. That's all.
If, regarding to this matter, there's anything computers can help with, it's understanding that allowing copies without restriction is not necessary as "evil" as some pretend to think.
This would have been considered obvious a few centuries ago. Art has not always been about the author.
War doesn't prove who's right, just who's left.
This is good work, but what is so special about it?
Using image analysis, "computers" these days do:
- Automatisation of drug discovery screening tests
- Diagnosis of skin cancer
- Detection of early breast cancers
- All sorts of QA in assembly lines
- And much much more, these are just examples you can find googling a bit.
Why is this news? If you go to any computer vision, image analysis or pattern recognition conference, you'll find many similar applications.
So, since this is purely a commercial program whose purpose is to provide a notional valuation based on association with celebrity, expect it to be extensively challenged. Too much is at stake. The art experts will soon weigh in there: the brush strokes being evaluated are actually those of the atelier assistants who did most of the work, the bits actually by the master defy analysis by a machine, and so on. Part of the value of the art market depends on gambling: finding the missed masterpiece, having a painting lose value owing to wrong attribution only to have the perceived value of the "real" artist increase as fashion changes. Anything that introduces apparent certainty will partly destroy the churning process that pushes art prices upwards, and no-one in the market wants that.
The price of art is as unrelated to the long-term assignment of aesthetic values as the price of CDs is unrelated to the actual merits of the performers. That's the sad reflection on our society.
Panurge has posted for the last time. Thanks for the positive moderations.
> John Cage sucks.
Yeah, I'd much rather use Raiden or Sub-Zero.
I've got a mind like a steel trap - it's got an animal's foot stuck in it.