Aesthetics in Digital Photography

5The Holy Grail of the Digital World: Artificial Intelligence

Is it permitted to suppose that a people whose eyes are accustomed to consider the results of a material science as the products of the beautiful won’t, at the end of a certain time, have singularly diminished its faculty of judging and feeling what is the most ethereal and most immaterial?

Charles BAUDELAIRE (1976)

The next three chapters will be dedicated to machine learning methods. The principle behind all the approaches that we will see in the following two chapters is based on the hypothesis that the rules that govern the appraisal of a photo can be derived from the observation of a large number of images judged by human experts. Implicitly, if we do not introduce any other mechanism, we can hypothesize that these rules are universal and timeless and we thus wholly adopt the objectivist framework.

Two groups of methods have been successively proposed. The first group was developed from classification algorithms, which were quite simple, with solid mathematical foundations. These used information selected from the image, most often chosen for the role attributed to them in the aesthetic judgment. These methods are said to be “primitive-based” or handcrafted. Then, from 2015 onward, these were slowly displaced by methods based on deep neural networks. These put the onus entirely on the power of the classification system (the deep network), omitting the step of selecting primitives. On the other hand, they required ...