AI helps ID paint chemistry of Berlin Wall murals
Briefly

Nondestructive techniques like Raman spectroscopy are crucial for analyzing pigments directly on-site, but they lack the precision of laboratory methods. This research aims to bridge that gap by employing machine learning to enhance the spectral data from handheld devices.
The analysis revealed that the paint fragments had predominantly brush-applied top layers, with an underlying layer likely serving as a preparatory surface. The presence of calcium and titanium was noted across all samples, signifying the base materials commonly used in the restoration efforts.
By developing their own mock-up samples using commercial acrylic paints, Armetta et al. sought to replicate the colors found on the Wall, demonstrating the importance of matching pigments for accurate restorations. The incorporation of SAPNet enabled the quantification of pigments with remarkable precision.
Ultimately, the integration of machine learning with Raman spectroscopy demonstrates a significant advancement in the field of art conservation, providing a comprehensive framework for identifying and quantifying color components in historical materials.
Read at Ars Technica
[
|
]