Teaching Computers to Appreciate Art
A Binary Mona Lisa
I recently came across an article about teaching computers to appreciate art – at least that was the title of the news story. The story actually dealt with a software program under development that would be able to analyze a work of art and associate it with an artist in its database. The software would also be used to spot art forgeries.
The program was created by Daniel Keren, a professor from the Department of Computer Science at the University of Haifa in Israel. The software works by subdividing known paintings by an artist into discrete blocks. Each of these blocks is then mathematically reduced to a formula. The multiple formulas can then be combined and compared with one another. The underlying program logic is that these sets of formulas will uniquely describe the artist that created the associated works of art.
Models were created for five artists – Vincent van Gogh, Rembrandt van Rijn, Salvador Dali, Rene Magritte, and Wassily Kandinsky – using 15 painting per artist. The program was then tested against another 15 paintings from each artist to see if the program could correctly associate a painting with an artist. The program was able to correctly associate a painting with its artist 86 percent of the time.
Over the course of the article it becomes clear that the program has a very long way to go in order for it to be considered truly functional. I am skeptical as to the program's chances of ever surpassing, or even equalling the judgments of the art specialist. While this program could develop to the point where it becomes a more useful tool, I can not see it ever reaching a state of reliability such that it becomes the arbiter of authenticity. I am even more skeptical of the ability to apply the program's technique to the analysis of digitally created art since digital brushwork does not have the same characteristics of traditional brushwork.
If you would like to know more about this program, see the article Teaching a Computer to Appreciate Art
Binary Mona Lisa
Since this post is about software “appreciating” art, I decided to create an image reflective of the subject. The image used to illustrate this blog entry is a digitally manipulated rendition of that most famous of paintings, the Mona Lisa, also known as La Gioconda, painted by Leonardo da Vinci slightly more than 500 years ago. The woman in the portrait is said to be the wife of Francesco del Giocondo.
Ad Astra, Jim