digital humanities; pose estimation; computer vision
Abstract :
[en] This paper introduces an approach for retrieving similar paintings in terms of poses of the characters depicted within them. The key contributions are a method to extract and normalize individual character poses, the development of several criteria to compare groups of poses across paintings, and the integration of these criteria into a unified ranking system to identify the most similar artworks for a given query. The proposed techniques are demonstrated on a corpus of religious paintings, showing their effectiveness in retrieving visually and semantically analogous artworks. The findings suggest that this methodology could prove helpful for extensive analyses of large image datasets.
