Reconstruction of a Hidden Page from Segmented Contours Using Deep Learning
Shunke Zhou, Quanli Pei · 2024
Historical documents are always too fragile to be opened, so the contents should be extracted carefully with less physical damage to the books. The significant improvements in X-ray scanning have made it possible to digitize and “virtual” these documents. A new technology involving a bunch of 3D X-ray images could help recognize the inner surfaces and create visualizations of written content [1]. The problem is extracting the book contents but avoiding opening them directly. This project aims to visualize the complete contents without secondary damage by using a CT scan to capture contour information and restructure the 3D surfaces of the books. This research will extract the edge or contour information from the segmentation stage and reconstruct 3D surfaces representing the book's individual pages. Once the pages have been reconstructed, text or image information will be mapped onto the surface of each page. This project aims to develop a computational framework for recovering surfaces from curves. The framework of this research provides a new technology for historians to better study historical books and protect the sealed books simultaneously. This study is very helpful for studying historical documents as it is a better way to extract contents from historical books with as little damage as possible than previous technologies.