Computers are able to understand art. Automatic translate

The task of correctly understanding and evaluating art still faces people. Can machines understand art? Programmers at Lawrence Technological University in Michigan, Lior Shamir, and Jane Tarakhovsky became interested in this issue. The results were astounding.

An algorithm has been developed that demonstrates that computers are able to "understand" art in a manner very close to the technique of artistic analysis, which art critics usually use in their work.

During the experiment, the results of which were published in the last issue of ACM Journal on Computing and Cultural Heritage, the researchers analyzed about 1000 paintings by 34 famous artists. A computer algorithm for analyzing the similarities between them was based solely on the visual content of the picture, without human intervention. Surprisingly, the computer divided artists into styles similar to the opinion of art historians.

The analysis showed that the computer is able to distinguish between classical realism and contemporary art styles. He automatically divided the artists into two groups, 18 classic artists and 16 contemporary artists. Within these two large groups, computer defined subgroups of artists that were part of the same artistic movement. For example, the computer automatically assigned to the Renaissance artists such as Raphael, Leonardo da Vinci and Michelangelo, very close to each other in style. Representatives of the Baroque style, artists Vermeer, Rubens and Rembrandt, were also assigned to the same group according to the algorithm. Automatic computer analysis is consistent with the point of view of art historians, who connect the styles of these three artists. In addition, the computer algorithm concluded that Gauguin and Cezanne, considered to be post-impressionists, have similar artistic styles, and also determined the similarities between the styles of Salvador Dali, Max Ernsat and Giorgio de Chirico, all of them, according to art historians, were part surreal art school. In general, the computer automatically performs analysis, which is largely consistent with the links between artists and art movements highlighted by art critics and critics.

The experiment showed that machines can surpass untrained people in the analysis of fine art. While the average person can only make a wide differentiation between modern art and classical realism, and with great difficulty finds the difference between the closely related schools of art, the computer easily separates the Early from the High Renaissance, mannerism from romanticism.

Within each experiment, 4027 parameters were selected for each picture by a computer, reflecting the content of the images, texture, expression and intensity of color, shape, nature of the borders between colors and shapes, etc. This made it possible to reflect a lot of aspects of visual perception, the statistical study of which made it possible to build models of kinship and differences between artistic styles, and further build a tree of similarity between different artists.

Anna Sidorova