LTU student shows that computers “understand” art
LTU student Jane Tarakhovsky showed, for the first time, that computers can match art historians in understanding and analysis of visual art.
In the experiment she let the computer analyze ~1000 paintings by 34 well-known painters, and let the computer automatically deduce the similarities between the artistic styles without using any information other than the visual content. The similarities were then visualized using a phylogeny (a tool normally used to visualize similarities between genomes of different species, but in this case was used to visualize the similarities between artistic styles). Surprisingly, the analysis of the computer was almost identical to the analysis of Art Historians.
For instance, the computer automatically placed the High Renaissance artists Raphael, Da Vinci, and Michelangelo very close to each other, and the Baroque painters Vermeer, Rubens and Rembrandt were placed by the algorithm in another cluster, indicating that the computer sensed that these painters share a common artistic style. The same was done by the computer for all 34 painters associated with schools of art such as Mannerism, Baroque, Rococo, Romanticism, Early, High and Northern Renaissance, Impressionism, Post and Neo Impressionism, Abstract Expressionism, Surrealism, and Fauvism. The similarity graph generated automatically by the computer showed that the computer analyzed the artistic styles in a fashion that is very similar to the way Art historians analyze influential links between painters and schools of art.
Her paper was published in the peer-reviewed journal “ACM Journal on Computing and Cultural Heritage”. It was also covered by the popular media around the world such as NBC News, Russia Today, and Times of India (the world’s most popular newspaper written in English).