Automatic Mosaic Digitalization: a Deep Learning approach to tessera segmentation

A deep learning approach successfully automates the segmentation of mosaic tesserae, significantly improving efficiency and accuracy over manual methods. This technique, applied to the mosaic floor of the Church of St. Stephen at Um er-Rasas (Kastrom Mefa'a), demonstrates high reliability in identifying non-homogeneous tesserae, a critical step for digital mosaic analysis. The method enables automatic cataloging, figure extraction, geolocalization, and semantic interpretation, addressing longstanding challenges in mosaic research. Experimental results on the collected dataset validate its effectiveness, offering a robust solution for large-scale mosaic digitization.

Felicetti, A., Albiero, A., Gabrielli, R., Pierdicca, R., Paolanti, M., Zingaretti, P., & Malinverni, E. S. (2018). Automatic Mosaic Digitalization: a Deep Learning approach to tessera segmentation. 2018 Metrology for Archaeology and Cultural Heritage (MetroArchaeo), 132–136. https://doi.org/10.1109/metroarchaeo43810.2018.13606