Schneider, StefanieSpringstein, MatthiasRahnama, JavadHüllermeier, EykeEwerth, RalphKohle, HubertusReussner, Ralf H.Koziolek, AnneHeinrich, Robert2021-01-272021-01-272021978-3-88579-701-2https://dl.gi.de/handle/20.500.12116/34717Due to the increasingly unmanageable number of art-historical inventories made available in digital form, methods that computationally arrange larger amounts of objects are becoming more important. The category of similarity, which is fundamental in all areas of art-historical description, gains new relevance in this context. In this paper, we propose a novel approach to the subject-specific classification of art-historical objects that utilizes expert-based attributes, i.e., significant figurative motifs. We evaluate our procedure on a concrete use case, representations of saints in the visual arts. A representative data set of saints images is collected and a semi-supervised learning technique applied to enrich the data set with neural style transfer as well as to improve the joint training of saints and their attributes. We show that this technique outperforms other approaches.enSemi-supervised LearningSemi-supervised Image ClassificationArt AnalysisDigital Humanities10.18420/inf2020_1271617-5468