4Algorithmic Approaches to “Calculate” Beauty
Science [...] represents the last remains of our relation to objectivity.
Luc FERRY (1990)
We will now look at research that proposes calculating the beauty of a photograph (or an image, or a painting) using a program rather than any human.
This research can be divided into two large, highly unequal groups: the first, which we discuss in this chapter, is devoted to algorithmic methods that propose comparing a photo with a calculation whose parameters are known and universal. These methods are thus heavily based on the objectivist approach. The second group works with methods based on machine learning using examples scored by a human observer. This methodology will be studied across three chapters. The first chapter examines the databases used for machine learning, as well as the expertise used as reference. The following chapter looks at the methods developed from 2000 to 2014, which use primitives taken from images, with rules implemented by classifiers acting on the primitives. The third chapter is dedicated to deep neural networks, which have been used since 2015.
4.1. First steps: C. Henry
The first contributions to the computer-assisted determination of beauty were due to Charles Henry (1885, 1891). He clearly noted that determining the exact physical-chemical changes correlated with aesthetic perceptions lay beyond the scope of the science of his time. He thus proposed a general framework to evaluate the links between arousal ...