Whether it’s a tree in the foreground or a ship off in the distance, artists put a lot of thought into the placement of each element on the canvas. Now, researchers at the Korea Advanced Institute of Science and Technology and Chungbuk National University in Korea have used a new method to find out how the composition of paintings changed over time in Western art. By digitally analyzing thousands of paintings, they discovered that the average placement of the horizon in landscape paintings has slightly shifted over the years.
The researchers collected 14,912 images of landscape paintings from the WikiArt and Web Gallery of Art databases, and wrote a program to interpret certain aspects of these digital images. The paintings dated from the Renaissance until modern times, and the question the team set out to answer was whether they could see patterns or trends in the way these landscapes were composed.
The composition of a painting roughly describes which element goes where, and how much of the canvas each part of the painting takes up. This isn’t unique to painting. Introductory photography classes often teach students the “rule of thirds” – a guideline that asks the photographer to mentally divide the image into thirds and place important elements on the imaginary horizontal and vertical lines. If a future art historian were to analyze all the photographs taken by every photographer who abides by this rule of thirds, and measured where the horizontal and vertical lines in the final photos were, they’d be able to see that pattern of thirds emerge.
That is roughly what the researchers did here. In a recent article in PNAS they describe how they designed an algorithm to identify areas in the images that corresponded to horizontal or vertical lines. They specifically looked at landscape images, because those very often have clearly pronounced horizontal and vertical lines in the horizon, buildings, trees, and cliffs.
The algorithm compared the colors of the individual pixels in each image to first find the most prominent horizontal or vertical line that divides the canvas in two. Then it moved on to find the line that divides the next-largest area in two, and so on.
Because the collection of images used in this study consisted entirely of landscape paintings, the most prominent dividing line in most of the images was the horizon. And when the researchers looked at the data from all these images that were divided by a clear horizontal horizon, they noticed that the horizon slightly shifted over the years. In paintings from the 16th century, the horizon was higher than in those from the late 17th century. Then, lower horizons were the norm until the mid 19th century, when they shifted up again.
Besides time-specific trends like the shifting horizon, the team also checked their data to see if they could find local trends. After all, certain art styles are often regional. But when it came to composition, the trends didn’t seem to follow any regional patterns. However, the researchers do point out in their study that they mainly looked at Western art, so this is not necessarily true for all art.
This type of analysis isn’t limited to paintings. The same techniques could be used to study composition patterns in films, typography or photography, for example. It could be a way to quickly assess a large number of pieces to find a pattern or trend, like this study did. But we can’t leave art analysis entirely to computers. It’s merely a possible starting point for further art historical analysis, because an algorithm can only find patterns, not understand the human intentions behind them.