Visual content classifier for cultural heritage repositories

Anonymous

Abstract: This work presents a novel approach for the automatic creation of an aligned image / text training set for the generation of descriptions of the visual content of artworks. To do this, we develop a classification tool based on a mix of heuristic rules and deep learning. This classifier is able to identify statements that describe visual art content, out of complex cultural heritage text that contains a mix of many other types of information on context, medium, author, etc. Our results are very promising when tested on texts from the Museo del Prado collections.

Paper Type: short