This project will develop algorithms and software tools for extracting the written text from volumetric, non-destructive scans of scrolls, which were carbonized in the Villa of the Papyri in Herculaneum as a result of the eruption of Mount Vesuvius in A.D. 79. The library from Herculaneum, which was first discovered in 1752, is of immense interest to scholars worldwide. The excavated collection contains more than 1800 carbonized papyrus scrolls, and is the only library known to have survived from classical antiquity. All attempts at physically opening the scrolls have created profound damage, and many scrolls (nearly 300) are still intact. This project will build a computational framework in order to "digitally unroll" two complete scrolls, producing a comprehensive, non-invasive solution to the technical challenges and prescribing a path for the scholarly assembly and publication of all inaccessible texts in the Herculaneum collection. This will definitively establish that volumetric, non-destructive imaging methods can reveal complete texts that were long thought lost. The algorithms developed in this project will set the stage for the interdisciplinary discovery of new works from antiquity, the scaling of these methods and algorithms to the complete Herculaneum collection, and the broader interest in volumetric analysis for cultural heritage.
The Herculaneum scrolls were written with carbon-based inks that create a challenging scenario for volumetric methods because density variations are very small. This work will overcome that problem using phase contrast tomography from a synchrotron beam in order to build high resolution volumetric models that show better contrast at ink boundaries. Because the internal structure of the Herculaneum rolls is complex, chaotic, and fully 3D, slice-based segmentation strategies fail when the surfaces turn into the dominant slice directions. Instead the segmentation approach will be based on registered multi-modal scans and a 3D structure tensor approach for building surface estimates from local multiscale volumetric operators and region-growing methods. The estimated papyrus surface segmentation will be locally re-positioned using optimization methods that improve the expected surface texture, and then digitally flattened to produce clear, flat views of the internal surfaces of the scrolls. This work will expand volumetric imaging toward a more widely used representational platform and will address fundamental technical challenges in segmentation, registration, representation, and visualization. The project will implement the developed segmentation and unwrapping algorithms as software tools so that non-scientists will be enabled to extract new images of text directly from volumetric data sets of scrolls. For further information, see the project website at http://vis.uky.edu.
