Data+ 2026 (Duke): Unveiling Hidden Inscriptions under a 700-Year-Old Painting
We develop advanced tools to recover hidden text from the reverse side of a 13th-century illuminated manuscript that has been glued to cardboard. By leveraging hyperspectral imaging and develop machine learning tools, our goal is to unveil secrets that have been obscured for 700 years.To achieve this, our study introduces an innovative pipeline that applies high-dimensional separation models to decipher text hidden beneath different pigments, relying on ideas of manifold learning in high and low dimensions. This approach integrates image processing, spectral data analysis, and NLP to bring these ancient words back to light.
PM: Keyu Li
Alex address a fundamental challenge in multimodal image registration for cultural heritage. Due to physical constraints, it is often impossible to capture an entire artwork in a single scan. To solve this, our framework mosaics Hyperspectral Imaging (HSI) data together using a high-resolution RGB image as a baseline guide. We then register various modalities, such as HSI and XRF, to the RGB image, while rigorously evaluating the accuracy of the alignment.
Anna's research focuses on applying novel optical techniques to problems in cultural heritage conservation. For example, applying pump-probe microscopy to differentiate and analyze dyes. Anna received her B.S. in Biochemistry from Elon University in 2022 and is currently pursuing a Ph.D. in Chemistry at Duke University.