A software reveals hitherto unknown influences between … – 21 Trends

id=”sep_para_1″ class=”sep_para access”

The task of sorting artwork is quite complex. When examining a painting, an expert can usually determine your style, genre, artist and the period to which it belongs. Art historians often go further, seeking influence and connections between artists, a task that is even more complicated.

So the possibility that a computer is able to classify paintings and find connections between them at first glance may seem laughable. Yet that is exactly what Babak Saleh and his team have succeeded in Rutgers University in the United States. Using some of the latest techniques of image processing and classification have automated a process that could take years done manually, to discover how great artists have influenced each other. They have even been able to detect influences of artists hitherto unrecognized.

According to an article in Physics arXiv Blog, the automated method is based on a new technique developed at Dartmouth College, USA, next to the Microsoft research center in Cambridge, UK, for classification According image containing visual concepts. These concepts range from simple description of the object, the color tones or more detailed descriptions.

The image analysis thus becomes a process of comparison of the words that describe them, for which there are well-established techniques.

id=”sep_para_2″ class=”sep_para access”

Visual Concepts

Saleh and his team applied this technique of machine vision over 1,700 paintings by 66 artists from 13 different exponents styles, framed in the period from the beginning of the fifteenth century to the twentieth. Parallel collated the resulting data with expert opinion on the influence of certain artists over others.

For each table, limited to 3,000 the number of concepts and points of interest generated, resulting in a list of descriptive words to consider as a kind of vector. The task is then to find similar vectors using natural language techniques and a machine learning algorithm.

The problem lies in the definition of influence, ie, determining if one considers that an artist influences another when one of his paintings has a strong similarity to another, or whether there should be a series of Similar paintings and, if so, how many. Seeking answers worked with different metrics. They ended up creating graphs with different metrics on each axis after plotting the position of each artist in this space to see how they grouped.

The results are interesting. In many cases, the algorithm clearly identifies influences and recognized by experts. It is the case of the Austrian painter Klimt, which the graphs show about Picasso and Braque found some left over.

The algorithm is also able to identify individual paintings that have influenced others. That tied Spanish Still Life: Sun and Shadow Pablo Picasso Man and violin Georges Braque, both painted in 1912 and recognized as the origin of Cubism. However, it also recognized the similarities between Old vineyard with peasant woman (1890) by Vincent van Gogh and The Farmhouse (1922) by Joan Miró, albeit very different style share landscapes and symbolism.

id=”sep_para_3″ class=”sep_para access”

Scoop

But most impressive was the link between The study of La Condamine Street (1870) by Frederic Bazille and The Barbershop Shuffleton (1950) by Norman Rockwell a connection never identified. “After researching many publications and websites, we conclude that this comparison has not been done before,” Saleh said.

“The paint may not be similar at first glance. However, a closer look reveals striking similarities in composition and objects are captured by our automated method,” detailed in the published study on the research, entitled Towards automated discovery of artistic influences.

And it’s both Rockwell painting of a barbershop seen through a window as her study Bazille heating stoves are on the right side; roughly where Bazille has a window, a door placed Rockwell; and composition of objects in both works creates a triangular space in the lower left corner.

Of course, researchers do not claim that this type of algorithm can relieve the site of an art historian. After all, the discovery of a link between paintings in this way is just the starting point for future research on the life and work of an artist.

“Our goal is not to get a final answer,” admits Saleh. Rather, it is used as a “tool” for art historians, help them do their job. However, it is fascinating to see how techniques Machine learning can shed new light on such a broad field as well studied art history.

In fact, is not the first machine vision technique applied to the study of art. Face recognition, for example, also be used to identify unknown characters in popular artwork, as stated above another article published in Tendencias21 .