Draw the border to answer the big questions of sorts
Algorithms have shown that the compositional structure of western landscape images changed “suspiciously” smoothly between AD 1500 and 2000, which may be due to a selection bias by art curators or in art historical literature by physicists from the Korea Advanced Institute of Science and Technology (KAIST) and colleagues report in the Procedure of the National Academy of Sciences (PNAS).
KAIST’s statistical physicist Hawoong Jeong worked with statisticians, digital analysts, and art historians in Korea, Estonia, and the United States to clarify whether computer algorithms can help resolve long-standing questions about design principles in landscape paintings such as the placement of the horizon and other key features.
Figure 1. The algorithm breaks down the image step by step based on the amount of information in each subsequent partition. Image credit: KAIST
“A fundamental question among art historians is whether works of art contain organizational principles that transcend culture and time, and if so, how these principles have evolved over time,” explains Jeong. “We have developed an information-theoretic approach with which the compositional relationship in landscape images can be captured, and we have found that the preferred compositional relationship has developed systematically over time.”
Digital versions of nearly 15,000 canonical landscape paintings from the Western Renaissance in the 16th century to the more recent contemporary art were executed using a computer algorithm. The algorithm gradually divides the image material into horizontal and vertical lines, depending on the amount of information in each subsequent partition. It enables scholars to evaluate how artists and different styles of art compose landscape graphics, in terms of the placement of key components of a piece, as well as the height or depth of the landscape’s horizon.
Image 2. The images were divided into four categories depending on how their compositions were divided by the first two lines developed by the algorithm. Image credit: KAIST
The scientists first analyzed the first two dividing lines identified by the algorithm in the paintings and found that they could be divided into four groups: an initial horizontal line followed by a second horizontal line (HH); an initial horizontal line followed by a second vertical line (HV); a vertical line followed by a horizontal line (VH); or a vertical line followed by a vertical line (VV) (see picture 1 and 2). They then looked at the categorizations over time.
They found that before the mid-nineteenth century, HV was the dominant compositional type, followed by HH, VH, and VV. The mid-nineteenth century then brought changes, with the HV compositional style losing popularity as the HH compositional style increased. The other two styles remained relatively stable.
The scientists also studied how the horizon line separating the sky and land has changed over time. In the 16th In the 19th century, the painting’s dominant horizon line was above the center of the canvas, but in the 17th century it gradually sloped down to the lower center of the canvas, where it remained until the mid-19th century. After that the horizon line gradually rose again.
Interestingly, the algorithm showed that these results were similar across cultures and artistic periods, even in periods dominated by a variety of different art styles. This similarity can therefore be a function of a distortion in the data set.
“Over the past few decades, art historians have prioritized the argument that the evolution of artistic expression has varied widely rather than offering a relatively smoother consensus story in Western art,” says Jeong. “This study serves as a reminder that the vast amount of data available may perpetuate serious biases.”
Next, the scientists want to expand their analyzes to more diverse works of art, as this particular dataset was ultimately Western and male biased. Future analyzes should also consider diagonal compositions in paintings, it is said.
Reference: “Analysis of the history of landscape art with information theory” by Byunghwee Lee, Min Kyung Seo, Daniel Kim, In-seob Shin, Maximilian Schich, Hawoong Jeong and Seung Kee Han, October 27, 2020, Procedure of the National Academy of Sciences.
DOI: 10.1073 / pnas.2011927117
This work was supported by the National Research Foundation (NRF) in Korea.
