A Deep Transfer Learning Approach for Writer Identification in Greek Papyri

Papyrology, the study of ancient texts on papyri, presents a significant challenge for scholars in identifying the writers, or scribes, responsible for these preserved texts. Traditionally, paleographers relied on qualitative methods for writer differentiation. However, recent years have seen the integration of computer-based tools, enhancing manual techniques by enabling automated measurement of parameters like letter dimensions, spacing, angles, and abbreviations. This evolution, digital palaeography, represents a transformative shift in palaeography, where digital technologies complement and enhance traditional manual methods. It combines advanced AI algorithms with high-resolution digital imagery, facilitating the extraction of unique manuscript features for writer classification. We characterize a writer's handwriting by leveraging different image processing methods and the ability of Deep Neural Networks (DNNs) to extract features from images automatically. This underscores the optimization of computational methods and digital imagery in advancing and supporting palaeographical analysis.