Jan 28th, 2008
What’s the color of that word?
If I say leaf, what color comes to your mind? In most likelihood, you’ll answer green for spring and summer and yellow or red for fall. That’s an easy one. Now, what if I tell you words like «daily » or « media », what color comes to your mind? A team of scientists from M.I.T. answered that question by giving every word of the English language a color. The result is available after the jump.
Furthermore, I use this orginal idea to search for the « color » of artists like Paul Gauguin and Marc Chagall.
A team of researchers from M.I.T. has used the Search Images fonction from Google to obtain 140 images for 53,464 nouns in the English vocabulary. The result is a collection of 7,527,697 images associated with a noun.
For some words, the resulting picture is easy to recognize while others require a bit more effort since it is blurred.
By using a tree-structured semantic hierarchy, the colored images were placed logically next to one another to create one huge image. As it is, two images were as close as the semantic link between them. For example, images associated with plants were in one area of the resulting picture, creating a green area easy to spot. The same can be said of human terms, creating a beige/brown area.
Even though the final image has some aesthetic value of its own, the main goal of the experiment was to explore how images from the Net could be used to train computers in performing the act of visual recognition.
Out of curiosity, I began to wonder what colors were associated with artistic terms. I searched a bit on the map and I found, in position 17121, the tile associated with the Pre-raphaelites.
The Pre-raphaelites were an English group of painters working in the 19th century who thought that art had gone bad since the days of Raphael (1483-1520). Their pieces mainly represent biblical or historical scenes. Of course, the human figure plays a key role in these types of representations.
Accordingly, it is not surprising that the main color associated with that word of the English language would resemble the human skin. Brown and beige tones are dominant in the tile.
What if I were to concentrate my efforts on a single artist? Did they find a dominant color associated with that artist? Could art historian use that research to further knowledge in that regard?
I began scrutinizing the image to find tiles associated with individual artists. And then I found one: Gauguin, in 17443. The colors associated with the artist, resulting from a selection of 140 images collected on the Web, seem to be linked to the Earth: brown, yellow, orange, dark green. This result is a bit surprising since the images used to create this composite tile show quite a diverse palette of colors. In fact, were is the blue of the tile represented in position 9?
Some of the art works of Gauguin represented are Manao Tupapau (L’esprit des morts veille), (1892, position 1), Autoportrait au Christ Jaune (1889-1890, position 12) and Cavaliers sur la plage (1902, position 13).
While I didn’t notice a huge difference in the resulting tile between the Pre-raphaelites and Gauguin, Marc Chagall created a bit of a surprise. I don’t find it surprising since Chagall’s work is malmost dream-inspired andvery colorful. Colors are varied and rich: pruple, yellow, green, red, blue… the entire spectrum is represented. Isn’t this like saying that Chagall doesn’t have a dominant color in his palette? What would Chagall’s specialists say about that, I wonder?
This makes me wonder about the methodology the researchers used to create this composite image. In all fairness, I must confess that I didn’t find the details of their method on their site.
The team of researchers say that they created the composite tile from a sample of 140 images returned from Google Images. However, if I type « Gauguin » in the Google and I look for images associated with that keyword, some pictures come up more often than others. This is easily understandable since masterpieces and well-known pictures are most likely associated with the artists than his lesser known pieces. How did the team of researchers select the pictures that created the tile associated with Gauguin? Can we say that that tile is the « color » of the artist?
Computer recognition of images is still in its infancy. It fascinates me to see the questions and the opportunities this type of researches could have for art history.
