Computers can only figure out a painting’s intricacies

LONDON - Computers can pretty well figure out the colour composition or aesthetics of paintings, but still lag behind humans in interpreting art.

How does one place an artwork in a particular artistic period? This is the question raised by scientists from the Laboratory of Graphics and Image in the University of Girona and the Max Planck Institute for Biological Cybernetics, Germany.

Researchers have shown that certain artificial vision algorithms (list of well defined instructions) mean a computer can be programmed to “understand” an image and differentiate between artistic styles based on low-level pictorial information.

Low-level pictorial information encompasses aspects such as brush thickness, the type of material and the composition of the palette of colours.

Human classification strategies, however, include medium and high-level concepts.

Medium-level information differentiates between certain objects and scenes appearing in a picture, as well as the type of painting (landscape, portrait, still life, etc), while high-level information takes into account the historical context and knowledge of the artists and artistic trends.

“It will never be possible to precisely determine mathematically an artistic period nor to measure the human response to a work of art, but we can look for trends,” Miquel Feixas, the study co-author said.

Researchers analysed various artificial vision algorithms used to classify art, and found that certain aesthetic measurements (calculating “the order” of the image based on analysing pixels and colour distribution), as well as the composition and diversity of the palette of colours, can be useful.

The team also worked with people with little knowledge of art, showing them more than 500 paintings done by artists from 11 artistic periods, said a release of SINC.

The participants were “surprisingly good” at linking the artworks with their corresponding artistic period, showing the high capacity of human perception.

These findings were published in Computers and Graphics.