Giuseppe De Gregorio
Postdoctoral researcher specializing in Computer Vision, Document Analysis, and Handwriting
Recognition, with a PhD in Information Engineering. My doctoral work centered on Word Spotting in
historical manuscripts, where I developed deep learning techniques for subword-level detection in
challenging, low-quality documents, a problem that demands robustness under extreme data scarcity
and visual noise.
My current research focuses on handwriting recognition in low-resource scenarios, working with
some of the most demanding document types available: Ancient Greek papyri, ciphered manuscripts,
and rare scripts. To tackle these challenges, I draw on self-supervised learning, synthetic data
generation, and in-context learning with vision-language models, methods that sit at the intersection
of classical computer vision and modern large-scale model capabilities.
I have also worked on layout analysis, character clustering, and retrieval-based approaches, giving me
a broad foundation across the document understanding pipeline. I have published at leading peerreviewed
venues (ICDAR, ICFHR) and co-organized tutorials and workshops at major conferences in
document analysis and machine learning.