Self-supervised learning for medieval handwriting identification: A case study from the Vatican Apostolic Library

In this paper, we consider the task of automatically identifying whether different parts medieval and modern manuscripts can be traced back to same copyist/scribe, a problem significant interest in paleography. Currently, application deep learning techniques context scribe recognition has been hindered by lack sufficiently large, labeled dataset, since labeling process is incredibly complex time-consuming. Here, propose first successful recent framework self-supervised field digital paleography, wherein pretrain convolutional neural network leveraging large amounts unlabeled manuscripts. To end, build novel dataset consisting both for copyist identification extracted from Vatican Apostolic Library. We show that fine-tuning model significantly outperforms other baselines, including common setup initializing general-domain features, or training scratch, also terms generalization power. Overall, our results reveal strong potential where data (i.e., digitized manuscripts) nowadays available, while scarcer. • Self-supervised improves handwriting on Self-supervision effectively addresses scarcity pretraining enhances new scribes. features outperform ones identification.