New artificial intelligence can complete the complex task of classifying fine art pieces better than human art historians.

Photo Credit: Medium

Photo Credit: Medium

Examining pieces of fine art is a pretty complicated process, which includes determining style, genre, artist and period. Art historians take this a step further by looking for artist influences. Babak Saleh and his team at Rutgers University have developed a computer that can do the job for art historians, successfully finding links between artists. 

According to Medium, the team used image processing and classifying techniques to automate the process of how some of the world's greatest artists have influenced each other. The computer was even able to discover influences that went unnoticed by art historians. 

Art experts usually approach this task by comparing artworks on number of factors, including space, texture, form, color, shape, and many others. They also look at the use of movement, harmony, variety, balance, contrast, pattern and proportion. Even aspects such as brush strokes, subject matter, historical context and meaning are taken into account. 

Saleh's team drew upon a new technique developed by Dartmouth College and Microsoft research for classifying pictures. Each image contains visual concepts called classemes and includes simple object descriptions, such as walking, shades of color, a tree and so on. Then, the process of comparing images becomes about comparing the descriptive words. 

The Rutgers team used this approach in classifying 1,700 paintings by 66 artists in 13 different styles. The time period of these art pieces range from the early 15th century to the late 20th century. They also collected expert opinions on which artists might have influenced others. 

For individual paintings, the concepts and points of interest were limited to 3,000. The process generates a list of descriptive words that can act as vectors. Then, the team has to look for similar vectors using natural language techniques and a machine learning algorithm. 

The artists' influences are harder to determine because the concept itself isn't an easy one. Saleh and his team experimented with various metrics and ended up creating two-dimensional graphs that display metrics on each axis then plot artists' positions on the graph to see how they cluster. 

In some cases, the algorithm was able to identify influences already found by art historians. It was also able to identify individual paintings that have influenced others. The algorithm was also able to link two paintings, Frederic Bazille’s Studio 9 Rue de la Condamine (1870) and Norman Rockwell’s Shuffleton’s Barber Shop (1950), which have not been previously linked by art historians.

“After browsing through many publications and websites, we concluded, to the best of our knowledge, that this comparison has not been made by an art historian before,” the research team said.

Saleh and his team do not think their algorithm could replace art historians, but could simply provide a starting point for further research.