Title: Learning of art style using AI and its evaluation based on psychological experiments

Authors: Mai Cong Hung; Ryohei Nakatsu; Naoko Tosa; Takashi Kusumi

Addresses: Osaka University, 1-1 Machikaneyama-cho, Toyonaka, Osaka, 560-0043, Japan ' Kyoto University, Yoshidahonmachi, Sakyo, Kyoto, 606-8501, Japan ' Kyoto University, Yoshidahonmachi, Sakyo, Kyoto, 606-8501, Japan ' Kyoto University, Yoshidahonmachi, Sakyo, Kyoto, 606-8501, Japan

Abstract: Generative adversarial networks (GANs) are AI technology that can achieve transformation between two image sets. Using GANs, the authors carried out a comparison among several artwork sets with four art styles: Western figurative painting set, Western abstract painting set, Chinese figurative painting set, and abstract image set created by one of the authors. The transformation from a flower photo set to each of these image sets was carried out using GAN, and four image sets, for which their original artworks and art genres were anonymised, were obtained. A psychological experiment was conducted by asking subjects to fill in questionnaires. By analysing the results, the authors found that abstract paintings and figurative paintings are judged to be different and also figurative paintings in the West and East were thought to be similar. These results show that AI can work as an analysis tool to investigate differences among artworks and art genres.

Keywords: generative adversarial networks; GANs; art genre; art history; style transfer; figurative art; abstract art.

DOI: 10.1504/IJART.2022.128444

International Journal of Arts and Technology, 2022 Vol.14 No.3, pp.171 - 191

Received: 02 Sep 2021
Accepted: 05 Jan 2022
Published online: 23 Jan 2023 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article