Machine Vision Reveals Previously Unknown Influences Between Great Artists

KentuckyFC writes Art experts look for influences between great masters by studying the artist's use of space, texture, form, shape, colour and so on. They may also consider the subject matter, brushstrokes, meaning, historical context and myriad other factors. So it's easy to imagine that today's primitive machine vision techniques have little to add. Not so. Using a new technique for classifying objects in images, a team of computer scientists and art experts have compared more than 1700 paintings from over 60 artists dating from the early 15th century to the late 20 the century. They've developed an algorithm that has used these classifications to find many well known influences between artists, such as the well known influence of Pablo Picasso and George Braque on the Austrian symbolist painter Gustav Klimt, the influence of the French romantic Delacroix on the French impressionist Bazille, the Norwegian painter Munch's influence on the German painter Beckmann and Degas' influence on Caillebotte. But the algorithm also discovered connections that art historians have never noticed (judge the comparisons for yourself). In particular, the algorithm points out that Norman Rockwell's Shuffleton's Barber Shop painted in 1950 is remarkably similar to Frederic Bazille's Studio 9 Rue de la Condamine painted 80 years before.

Comments

Copyright harassment

[tepples]

How long until dead artists' heirs latch onto techniques like this to prove "substantial similarity" as part of a copyright suit to try to squeeze money out of working artists?

Re:Copyright harassment

[tepples]

That isn't "Melancholy Elephants" by Spider Robinson, is it?

But how long until computers generate good art?

[penguinoid]

I suppose the only way people will quit caring more about which artist drew something than how it looks, is if we replace artists with computers.

Influence vs. similarity

[kruach aum]

What the computer can do is point out what is similar. Whether the similarity is an example of influence then needs to be established with further evidence.

Re:Influence vs. similarity

[Chrisq]

What the computer can do is point out what is similar. Whether the similarity is an example of influence then needs to be established with further evidence.

Of course even if there is influence it may not be direct. Both artists could have been influenced by another earlier artist, or the second one influenced by an intermediate one, or there could be a whole complex tree of influences

Re:Influence vs. similarity

[wisnoskij]

Or they could of both been influenced by using the same tools, making the same paints in the same way, or being in the same environment.

Re:Influence vs. similarity

[globaljustin]

Exaclty...

when i used to do music reviews I'd interview artists and they'd talk about similarity of sound mostly as a function of the type of instrument or accoutrement they used...

Re:Influence vs. similarity

[plover]

The human can only do that if both pictures come to his attention. But there is so much out there that it's almost impossible for someone to be familiar with every piece to the extent they'd be able to recognize them. The computer has infinite patience, it can attend to vast quantities of the most minute details, it has a catalog that doesn't fade with time, and the ability to re-run increasingly sophisticated algorithms as new ideas are brought to bear.

For example, Rockwell's barber shop and Bazille's studio share a few subjects in a few common locations, but it's hard to look at them and say "there was an artistic influence." Rockwell was noted for realistic depictions of idyllic Americana, so any influence there would likely have been the architecture of the setting and the choices of overall composition and balance. Choosing to include a group of three people, an unoccupied chair, and a wood stove, does not seem to imply much more than coincidence. But if you weren't comparing every item in the catalog with every other item in the catalog, you might not have bothered to notice at all.

Which brings us to the real question: how would knowing the answer (or even asking the question) make a difference to the world?

Re:Influence vs. similarity

[plover]

Actually, the more I look at the Rockwell and the Bazille, the more sophisticated the results of the comparison appear to be. You've got a group of men, off in the background, engaging in a conversation that you are not able to hear. They're the subjects of the piece, but you don't see much of them, you can't hear what they're saying, and what they're talking about is partially obscured. You assume that because they're invited to the back room of the barbershop that they're more than just customers, similarly the men discussing the painting appear to share a common interest. The stoves suggest that a warmth exists, and that the people are physically comfortable in both places. The empty foreground spaces indicate a purpose that's going partially unused at the moment. The chairs give an identity to each place: the barbershop chair helps you understand that it's a shop, and because no one is sitting in it, you realize that a discussion other than banal haircut chatter about the ballgame is going on. The empty salon chair lets you know that the studio is underutilized - maybe this is a showing of unpopular works?

I still think that the paintings are likely unrelated to each other, but it seems that both artists were thinking similar thoughts when they chose to paint these. And that's the sophistication of the algorithm.

Similarities seem kind of tenuous

[timrod]

I looked at the Rockwell/Bazille comparison, and they don't really seem all that similar - they have three similar elements (stove, chair, and window) but those seem coincidental more than anything. The window in Rockwell's piece, for instance, is small and rectangular while the one in Bazille's is huge and arched. The chair in the Rockwell piece is actually barely identifiable as a chair at first glance, whereas the one in the Bazille piece is immediately recognizable as a wooden chair. They're also three objects that are likely to be close to one another. For instance, my aunt heats with wood and has a stove roughly the same distance from a window as in the Rockwell and Bazille pictures, and if I remember right even has a wooden chair in the same room. I think all this proves is that people tend to put their stoves in rooms with windows and chairs.

Re:Similarities seem kind of tenuous

[oodaloop]

I suppose that's why they say this then:

...Saleh and co do not claim that this kind of algorithm can take the place of an art historian. After all, the discovery of a link between paintings in this way is just the starting point for further research about an artist’s life and work.

I don't see it

[Nidi62]

So...because Rockwell and Bazille's paintings both have windows, people, a chair, and a stove they are influenced by each other? All of these are common things that you would expect in any building in the late 1800s and mid-50s (note the age of the building implies it is not new construction at that time and would definitely still rely on a stove for heating). I guess they are trying to argue that the placement of the items is the connection? Barbershops always have their chairs on one side near the wall, and people tend to put chairs near walls and objects as well, not in the middle of the floor. The right angle formed by the wall and floor and then the pane in the window seems a bit of a stretch, since wouldn't any painting of a man-made structure include right angles at some point?

I guess I just don't "get it"

Re:I don't see it

[c]

I guess they are trying to argue that the placement of the items is the connection?

Pretty much. I suspect this is one of those situations where "correlation != causality" is an appropriate comment.

Re:I don't see it

[Skewray]

I guess they are trying to argue that the placement of the items is the connection?

Pretty much. I suspect this is one of those situations where "correlation != causality" is an appropriate comment.

I would say instead that, given a sufficiently large enough data set, patterns and correlations are bound to appear. The likelihood that thousands of paintings were analyzed in this way and no matches were found, purely on a random basis, is very small.

Re:I don't see it

[jedidiah]

Just like with evolution, two similar things don't have to be parent and child. They can be siblings or cousins sharing a common origin.

Rockwell and Bazille likely share common influences including the general culture at large.

This is the sort of thing that should be obvious to anyone that's ever been in any serious kind of art museum. Art documents the culture that created it. You can easily see how that changes over time.

A number of these cultural transitions are really quite dramatic.

M.C. Escher

[The Grim Reefer]

Interestingly when M.C. Escher paintings were analyzed it kept returning "divide by zero errors". Upon further examination, it was discovered that it was claiming "divide by the letter "O" errors" and not the number zero.

Re:M.C. Escher

[kruach aum]

I started thinking about Escher too when I read this, and maybe it's because my knowledge of art history is simply inadequate, but I couldn't think of any artist influenced by Escher other than the creator of the Berserk manga, Kentaro Miura. And even there "influence" may be too strong, as it's more that Miura visually references Escher a couple of times.

Re: artist's use of space

[i.r.id10t]

Dunno but it may help to keep him on ice ice baby if he is under pressure a lot ....

Similiarity being the key

[s.petry]

Art is taught and can be learned, I study this as well as everything else I can find to learn (yes, even at a University). What an artist learns is how to move a persons eye, and how to make aesthetically pleasing art, amongst other things.

As an example, If you draw an X in the center of the canvas and maintain the lines in the painting, people's eyes will be drawn to the center. Numerous Xs will have numerous focal points. Great artists know this, and obscure the lines so it's harder for people to notice. Flat lines are obvious and can be boring.

Layering paints is another one that people can be taught, but can learn on their own. I don't have to teach you the brush strokes required, you would figure it out. If you press a brush on canvas and drag the brush, paint moves in the same way as the brush. Layering requires a touch lift technique (or waiting until paint is dry). We don't need to have the same influences to figure this out, just a little bit of experience with the medium.

In "teaching" art I may be able to hasten your learning curve, but these are two concepts that you would surely figure out on your own. Unless I know you were classically trained I can't claim to know how you were influenced by looking at simply technique.

Like you said, I have no issue with them claiming that they can see what's "similar". Unless an artist admits to being trained by so-and-so or influenced by so-and-so, I don't buy into a piece of software being able to do this just by looking at technique.

Correlation does not imply causation.

[Thatto]

An algorithm scanned through images of works of art, and identified similarities. And that means what?

Re:"program" = cattle prod

[penguinoid]

humans are not "just machines" because we can **choose to program ourselves and formulate/test hypothesis that we communicate/share/compare with others**

So can computers. Computers can run arbitrary code. Computers can generate arbitrary code. Humans are much more limited in their ability to chose how their mind functions (which is ironically why many people think humans have free will).