MSI+

Context

Palimpsests are historical manuscripts in which earlier text has been deliberately erased or scraped away to make room for new writing, a common practice in antiquity and the Middle Ages due to the high cost and scarcity of writing materials such as parchment. This reuse often leaves faint traces of the original (undertext) beneath the later (overtext), creating complex, overlapping layers that are extremely difficult to decipher with the naked eye or conventional imaging. Paleographers and historians face major challenges in recovering these hidden layers, as natural degradation, including ink fading, parchment discoloration, surface damage, and domain-specific noise, further obscures the original content. Traditional methods struggle to separate the intertwined layers non-invasively while preserving paleographic authenticity, limiting access to invaluable cultural and historical knowledge embedded in these manuscripts.

Objective

The primary objective of the MSI+ project is to develop an advanced generative AI framework that integrates multispectral imaging (MSI) data across ultraviolet, visible, and near-infrared bands to enhance and reconstruct hidden undertext in palimpsests. By leveraging modern AI techniques, the project aims to produce clearer, more readable images of the original erased text and enable more accurate subsequent text extraction. The approach focuses on transforming raw, degraded multispectral data into high-fidelity enhanced representations suitable for paleographic analysis, while emphasizing data efficiency and robustness in the face of limited real-world training samples typical of historical manuscripts.

Knowledge Gain

This project seeks to bridge the longstanding gap between black-box deep learning models and the interpretability requirements of the interpretive humanities. By incorporating domain-informed inductive biases and physically grounded representations of spectral properties, MSI+ aims to make AI-driven analysis more transparent and verifiable by experts in manuscript studies. The resulting methodological advances will contribute to computational paleography and digital humanities, offering new tools for non-invasive cultural heritage preservation. Ultimately, the work will facilitate broader access to previously inaccessible historical texts, supporting global efforts to safeguard and study our shared written heritage in a sustainable and scientifically rigorous manner.