How X-rays, geometry, and machine learning are reopening the Herculaneum scrolls
When Vesuvius destroyed Herculaneum in 79 CE, it preserved a library in a nearly unreadable form: papyrus scrolls burned, crushed, and sealed into brittle carbon. The central challenge has always been how to read these scrolls without physically opening and destroying them.
The scrolls are not simply closed books. They are tangled three-dimensional structures of carbonized papyrus, and in many cases the ink is carbon-based too, which makes ordinary contrast between writing and support extremely weak.
High-resolution X-ray CT, including phase-contrast imaging, changed the problem by letting researchers capture the internal volume of unopened rolls. But seeing the volume is only the beginning, because the text-bearing surfaces are curled, crushed, nested, and distorted.
Virtual unwrapping is the geometric heart of the work. Software must identify papyrus layers in the 3D scan, trace the sheet-like surfaces, and flatten them into readable two-dimensional images without inventing false structure.
Machine learning helps most when it is treated as a detector of faint physical evidence, not as a translator of ancient Greek. Models can be trained on fragments where visible ink aligns with CT data, then used to mark places where hidden ink may exist.
