Using computers to better understand art

Featured on theconversation.com

How do humans interpret and understand art? The nature of artistic style, seemingly abstract and intuitive, is the subject of ongoing debate within art history and the philosophy of art.

When we talk about paintings, artistic style can refer to image features like the brushstrokes, contour and distribution of colors that painters employ, often implicitly, to construct their works. An artist’s style helps convey meaning and intent, and affects the aesthetic experience a user has when interacting with that artwork. Style also helps us identify and sometimes categorize their work, often placing it in the context of a specific period or place.

A new field of research aims to deepen, and even quantify, our understanding of this intangible quality. Inherently interdisciplinary, visual stylometry uses computational and statistical methods to calculate and compare these underlying image features in ways humans never could before. Instead of relying only on what our senses perceive, we can use these mathematical techniques to discover novel insights into artists and artworks.

A new way to see paintings

Quantifying artistic style can help us trace the cultural history of art as schools and artists influence each other through time, as well as authenticate unknown artworks or suspected forgeries and even attribute works that could be by more than one artist to a best matching artist. It can also show us how an artist’s style and approach changes over the course of a career.

Computer analysis of even previously well-studied images can yield new relationships that aren’t necessarily apparent to people, such as Gaugin’s printmaking methods. In fact, these techniques could actually help us discover how humans perceive artworks.

Art scholars believe that a strong indicator of an artist’s style is the use of color and how it varies across the different parts of a painting. Digital tools can aid this analysis.

For example, we can start by digitizing a sample artwork, such as Albert Bierstadt’s “The Morteratsch Glacier, Upper Engadine Valley, Pontresina,” 1885, from the Brooklyn Museum.

Scanning the image breaks it down into individual pixels with numeric values for how much red, green and blue is in each tiny section of the painting. Calculating the difference in those values between each pixel and the others near it, throughout the painting, shows us how these tonal features vary across the work. We can then represent those values graphically, giving us another view of the painting:

This can help us start to categorize the style of an artist as using greater or fewer textural components, for example. When we did this as part of an analysis of many paintings in the Impressionist and Hudson River schools, our system could sort each painting by school based on its tonal distribution.

Click here to read the full article.