AI news ·
Machine learning method suggests El Greco painted more of The Baptism of Christ than previously thought
A machine-learning tool developed at Case Western Reserve University finds that El Greco's *The Baptism of Christ* was painted mainly by El Greco himself, not his workshop. The method analyzes brushwork in 1cm patches across high-resolution scans.
Machine-Learning Method Revises Attribution of El Greco Altarpiece
Researchers at Case Western Reserve University developed a computer-vision technique that challenges long-held assumptions about workshop contributions to a 16th-century El Greco painting. The method, published in Science Advances, suggests The Baptism of Christ was painted primarily by El Greco himself, not substantially by his workshop as art historians have believed.
The team created PATCH (pairwise assignment training for classifying heterogeneity), which analyzes 1-centimeter-square patches of paint and brushwork across high-resolution scans. By comparing texture and brushwork consistency, the algorithm produces spatial maps showing where different hands likely worked on a canvas.
How the method works
PATCH trains on paintings known to be single-artist works, learning the normal variation in an individual artist's technique. It then compares patches from test paintings and assigns them probabilistic scores: same artist, different artist, or group origin.
- Analyzes brushwork and paint texture at patch scale
- Produces spatial heterogeneity maps across entire paintings
- Generates probabilistic assignments rather than binary yes-or-no conclusions
- Integrates with archival research, pigment analysis, and expert annotation
The researchers framed PATCH as a complement to existing methods, not a replacement. "PATCH has the capability to make a substantial contribution to research as a complement to existing art historical methods," they said.
Practical application in provenance work
Attribution disputes often hinge on connoisseurship-the trained eye of experts-combined with archival records and materials analysis. PATCH adds quantitative evidence to these conversations, giving conservators and historians empirical data to evaluate claims about who painted what portions of a work.
The approach reflects a broader trend: applying high-resolution imaging and machine learning to material culture questions where stakes are high and expert disagreement common.
What remains uncertain
The method's generality is unproven. Validation across other artists, different workshop periods, and additional imaging types will determine whether PATCH works reliably beyond El Greco's oeuvre. Researchers plan to test the technique on curated ground-truth datasets and integrate it with spectral analysis of pigments.
Until those steps are complete, PATCH's impact will remain confined to specialized art-historical research rather than broader museum practice.
Learn more about AI for Science & Research and how machine learning supports evidence-based inquiry across disciplines.