Creating Digital Estampages of Sri Lankan stone inscriptions using Multispectral Imaging

Abstract

Stone inscriptions are of immense historical and cultural importance, but are progressively becoming illegible due to natural weathering, erosion and vandalism. This paper presents a novel approach to enhance the readability of stone inscriptions using multispectral imaging. Traditional methods such as estampages are weather-dependent and prone to human error. The proposed multispectral imaging based system combines advanced techniques such as spectral enhancement based on Kubelka–Munk theory, multispectral band selection, and deep learning–based preprocessing models (U-Net, ResNet) to significantly improve legibility, providing a non-invasive and efficient method for documenting and preserving ancient inscriptions. Results demonstrate that our U Net–based preprocessing pipeline achieved a Peak Signal-to Noise Ratio (PSNR) of 15.33 dB and reduced Fr´echet Inception Distance (FID) to 518, while expert evaluation showed a 58.7% preference for digital estampages over traditional methods. This demonstrate the system’s potential for integration into existing archaeological workflows, improving both accuracy and time efficiency in inscription documentation.