人工智能在文档修复和手稿恢复中的应用

The Vesuvius Challenge uses X-ray CT scans and machine learning to read carbonized Herculaneum scrolls without physically unrolling them. The pipeline reconstructs scroll geometry, segments papyrus layers, flattens their surfaces, and detects ink-related texture; carbon ink and papyrus can be difficult to distinguish because their densities are similar. Ithaca is a deep neural network for restoring missing text and estimating the geographic and chronological attribution of ancient Greek inscriptions; its outputs are candidate interpretations for historians. Handwritten text recognition (HTR) transcribes handwriting into searchable text, while image-enhancement systems address stains, fading, and other degradations. Multispectral imaging captures a document under multiple wavelength bands, including infrared; comparing bands can expose ink or degradation not visible in ordinary color images. A study of historical document restoration used infrared information to isolate degradation and reconstruct a cleaned color image; it does not establish AI reading of erased palimpsests. Each recovery is evidence-informed and should retain access to the original scan.