An AI-Assisted Classification-Driven Approach for Enhancement and Analysis of Palimpsest Readability

Abstract

Palimpsests are ancient manuscripts in which the original text has been partially erased and overwritten by a later one, which makes them very difficult to decipher. This study examines the possibilities for analyzing, evaluating, and classifying the legibility of such manuscripts using modern image processing and machine learning algorithms, and proposes approaches for improving it. The method investigated by the team involves feature extraction, a quantitative readability metric and automatic classification using algorithms such as PCA and K-Means. These techniques allow images to be grouped according to their condition. Building on this framework, advanced adaptive processing and classification methods are applied to improve the visualization of text layers. Our research is based on images captured in 2026 at the National Library in Sofia, as part of work on the 13th-century Dragotin Miney. The results showed improved legibility and more efficient preparation for subsequent transcription by museum staff. The proposed framework is intended to support museum specialists during the interpretation of severely degraded manuscript fragments under real archival conditions.