Reconstruction of the Baronci altarpiece by Raphael. The top two paintings, the Virgin Mary (left) and God the Father (right), are analyzed in this work. Both paintings belong to the collection of the Museo di Capodimonte (Napoli). At the bottom left is the Angel holding a phylactery (Louvre, Paris) and at the bottom right the image of the Angel (Pinacotheca Tosio Martinengo, Brescia). Credit: Danilo Pavone, ISPC-CNR, Catania
A team of chemists and AI researchers from CNR, Istituto di Scienze del Patrimonio Culturale, has developed an AI model capable of determining the chemical composition of paints used to make classic paintings.
In their article published in Scientific advancesthe group describes how they developed their AI model and trained it using datasets containing information on 500,000 synthetic spectra, representing 57 pigments and related compounds.
Maintaining and/or restoring old paintings, especially those of great value, is both an art and a science. Specialists have training in many fields, ranging from chemistry to botany and history. Due to the high value of these works of art, new techniques are sought to better understand the nature of a given painting before a restoration effort is undertaken.
A major area of interest is the chemical composition of the paints used by the artist. If the wrong chemicals are applied, reactions can occur, leading to paint degradation or even ruining an ancient masterpiece. In this new effort, the research team put artificial intelligence to the test.
High resolution MA-XRF elemental distribution maps of God the Father details. (A) Visible image of the scanned area. (B) Elemental map of Pb-L. (C) Elemental maps of Hg-L. (D) Composite red-green-blue (RGB) image of the elemental distribution maps of Hg-L, Fe-K, and Cu-K. The image provides insight into Raphael’s painting technique. (E) The elemental distribution maps of Pb-M. AI/ML estimates are lower than the baseline. The network, however, provides a less noisy output. (F) Elemental distribution maps of SK. The AI/ML image is predicted, while the reference, due to low net sulfur counts and energetic overlap of SK with Pb-M, is too noisy to interpret. Credit: Photo (A) by Danilo Pavone, ISPC-CNR, Catania; Scientific advances (2024). DOI: 10.1126/sciadv.adp6234
To better understand the chemicals that make up a given paint, experts use X-ray fluorescence: X-ray imaging is performed non-invasively and yields detailed elemental compositions associated with the paint used on a given paint. . Unfortunately, the fact that artists mix pigments to achieve the desired color makes it more difficult to identify individual paintings.
Attempting to determine the chemicals contained in such mixtures often involves educated guesswork, which leads to errors. To reduce these errors, researchers developed an AI model that can receive macro-X-ray fluorescence (MA-XRF) data sets, analyze them, and then print the chemicals found in all oils used to create a given painting. The model was trained using a dataset containing information on 500,000 synthetic spectra.
Once the model was completed and the first tests carried out, the research team gave it a more realistic test by asking it to identify the chemicals present in the oils used to create two paintings made by the artist Raphael between 1501 and 1502.
Both have been extensively studied and tested using other methods, meaning their chemical components have been previously identified. The research team found that the model was able to correctly identify chemicals including lead in white paint, mercury in red paint, and copper in green paint.
More information:
Zdenek Preisler et al, Deep learning for enhanced spectral analysis of MA-XRF datasets of paintings, Scientific advances (2024). DOI: 10.1126/sciadv.adp6234
© 2024 Science X Network
Quote: Using AI to determine the chemical composition of paints used in conventional paints (October 5, 2024) retrieved October 5, 2024 from
This document is subject to copyright. Except for fair use for private study or research purposes, no part may be reproduced without written permission. The content is provided for informational purposes only.
