What if there was a tool that could tell you which artists in art history are most like your artwork?
Upload 20 images of your artwork to this tool. Based on your art, a list of artists is generated most similar to your artwork.
How would this work?
The program would use machine learning to compare each of your images against a deep library of artwork by other artists. The program finds the closest match for each of your artworks.
For each of your 20 artworks, you might end up with a list like this:
- Image 1 most like Georges Braque
- Image 2 most like Juan Gris
- Image 3 most like Georges Braque
- Image 4 most like Jean Metzinger
- Image 5 most like Juan Gris
- Image 6 most like Georges Braque
- Image 7 most like Pablo Picasso
- … so on
Over the course of the 20 images, the program would find the artist who appears most often as a match for each of your artworks. In this example, Georges Braque would be the most similar artist. 2nd and 3rd most similar would also be displayed.
The value for artists
This would be of tremendous value to artists—to know which artists in history have done work like yours.
A valuable tip was given to me when I was in high school. At an art compeition, one of the judges commented my painting was like Edward Hopper. As a high schooler, I was only beginning to learn about artists, so I didn’t know about Hopper. Thus the Hopper tip was fantastic! I dug into his work and found out that I absolutely loved that I did something that was remotely close to a Hopper.
This would be a great tool for experienced painters too. Even if you are set in your artistic style, it’s great to see how others have a slightly different take on a particular style.
Yeah, in some cases, an experienced artist will most likely know of many artists similar their style, like an Edward Hopper. But with a deep database of art, there might be some lesser known artists that will emerge.
Controls over parameters in search
It would also be great to allow the artist using this tool to fine tune the preferences. Let’s say you only want to see artists who use a similar color palette as yours. Put the weight on color palette. More interested in texture? Put the weight on texture.
One of the frustrating aspects of AI and art is the opaqueness of the method. Give the user some control over the criteria, and that opens up some of the understanding of the connections.
In addition to populating the database with famous art, it would be fascinating to include every day artists from today too. Copyright issues will emerge of course with using other people’s artwork. I imagine a project like this will only be able to use artwork in the open domain, thus limiting the recommendations.
Nevertheless, I really like this idea.
Reverse Google Image search
Actually, there is a system that exists pretty close to this. Reverse Google Image search.
Upload your image to Google Image Search. Google will automatically try to guess what the image is about, and put a keyword in the search box. Change that keyword to an art website with a large database. I tend to use the Art Institute’s website, artic.edu. Thus, in the search box, I would remove Google’s generated keyword, and put in: site:artic.edu.
Let’s try this with my high school painting.
Uploading this painting to Google images, the generated keyword is “gentleman”.
Ok, that’s kinda cool, I guess? Gentleman does describe the painting. But we want it to search just the Art Institute’s site. Change “gentleman” to “site:artic.edu“
The first result is the Holland artist Samuel Dircksz Hoogstraten’s “Boy Looking through the Window” from the late 1640s. Kinda cool.
The second result is Dutch artist Evert Pieters’ “A Family Meal” from the 1890s.
Just for reference, here’s my painting again:
- James Whistler (American, 1834–1903)
- Frank Duveneck (American, 1848–1919)
- Walter Sickert (English, 1860-1942)
- Karel van der Pluym (Dutch, 1625-1672)
- Lovis Corinth (German, 1858-1925)
- Edward Steichen (American photographer, 1879-1973)
- Julius Gari Melchers (American, 1860–1932)
- Edouard Manet (French, 1832-1883)
Where’s Edward Hopper?
Hopper doesn’t seem to be in any of the top results. Let’s just change the search to be Edward Hopper. (now we aren’t searching artic.edu, but any Edward Hopper on the internet)
Boom, The first four results are all the same painting! “Room in New York” (1932). Very fitting. Man sitting in a chair. White sleeves, vest, tie. Door in the background.
Automating Reverse Google image search
Do this with a number of your artworks, and you’ll generate an interesting list of similar artists. Although it takes a while to manually click through each image match; and record the name of the artist.
Imagine if this was all automated in some way. Select 20 of your own images, Google does the matching, then you get a list of the top ten artists for each painting. The results also tell you which artist appears the most. It might also be interesting to see which style and time period appear frequently. In this case, for my painting, it would be the Dutch Golden Age.
I’ve started to delve a little bit into the world of Machine Learning (and Artificial Intelligence), maybe this tool might be something that comes into reality!