A machine as an art historian and a connoisseur of paintings?

In the past few years, thanks to neural network technology and computer learning, machines have been able to outdo humans in many pattern recognition tasks. Now is the time to analyze the works of art. Babak Saleh and Ahmed Elgammal of Rutgers University (NJ, USA) are working on an algorithm by which a computer can recognize artists and the style in which the work has been painted – something that even humans have trouble with.

A bit of statistics – Saleh and Elgammal collected digitized versions of more than 80,000 paintings from the brush of over a thousand artists, from fifteen different time periods. All artworks are assigned categories such as interiors, landscapes, portraits, etc., as well as a specific painting style; A total of 27 were distinguished, each of which is represented by approximately 1,500 works. Researchers presented the machine with sets of images and taught individual algorithms to find certain features in the analyzed paintings – from the simple determination of the predominant colors to more complicated recognition of the presented objects. From this data, the computer created a multidimensional description for each of the images. The scientists then tested the algorithms, giving them a completely new set for analysis, with which the machine had not yet come into contact. The results are impressive, because artificial intelligence was able to correctly identify the author of the painting in 60% of cases, and the style – in 45%.

Po lewej: Arshile Gorky – „The liver is the cock’s comb” (1944r.); po prawej: Jackson Pollock – „Number 8” (1949r.)

The machine finds connections between painting styles, although it still sometimes confuses abstract expressionism and its subgenre, gesture painting (consisting in accidental splashing of paint on canvas and unrestricted brush movements; Jackson Pollock is a well-known representative of the trend). The similarities with which the algorithms deal very well is capturing the relationship between expressionism and fauvism (fauvism is often perceived as a fraction of expressionism, characterized by extremely vivid, aggressive colors). The computer also correctly combines mannerism with styles in Renaissance art – mannerism is a Renaissance trend, characterized, among others, by sublime poses of the presented figures and exaggerated emphasis on beauty by violating the body proportions (for example, unnatural elongation of the limbs)

Monet_Pissaro — *Left: Claude Monet – “The Summer, Poppy Field” (1875); on the right: Camille Pissarro – “Autumn morning at Eragny” (1897)*

Artificial intelligence still has a problem where the differences are very fluid. This is the case with the works of Camille Pissarro and Claude Monet; only after delving into the history of both gentlemen, we discover that they worked in France at the turn of the 19th and 20th centuries and attended the Académie Suisse in Paris together. They were good friends, they shared their experiences and influenced each other’s art, hence many similarities between their works. Another example is the machine’s confusion between Monet’s paintings and the American impressionist, Childe Hassam, who was fascinated by French Impressionism. Monet’s works were the strongest inspiration for his work.

Read This Now: We know the Internet access rates of the Starlink Premium service from SpaceX

These nuances are known to art historians, although it took decades to accumulate and systematize this knowledge; neural networks take several months to learn. There is a chance that computers will shed some new light on the fine arts. One tool, an application for finding images with similar characteristics, can inspire historians to search for relationships and mutual influences between works or artists that have hitherto remained unnoticed.

Source: TechnologyReview