AI Detects Art Forgeries with 96% Accuracy Using 3D Imaging

Researchers at Case Western Reserve University in Cleveland have developed an AI system that detects art forgeries with up to 96% accuracy using chromatic confocal optical profilometry, a 3D imaging technique that analyzes brushstroke surface textures like fingerprints. The study, published in Heritage Science in November 2021, was led by physicist Kenneth Singer with a team of art historians and computer scientists. They used convolutional neural networks trained on four identical yellow flower paintings created by Cleveland Institute of Art students. The method identifies unique brushstroke characteristics at microscopic scales, even distinguishing individual bristle marks. It outperforms existing AI authentication methods like those used by Swiss company Art Recognition, which achieves 85% accuracy but requires hundreds of high-resolution images. The Case Western team collaborated with Madrid-based Factum Arte to test the technique on El Greco's Portrait of Juan Pardo de Tavera (1609), a painting damaged and restored after the Spanish Civil War. The AI successfully identified areas where restorers had intervened, distinguishing them from El Greco's original hand. The researchers plan to extend the method to detect multiple artists' contributions on a single canvas, potentially revealing which workshop assistants worked on Renaissance commissions versus the master's own hand. The technology works on oil paintings, acrylics, drawings, watercolors, and sculptures.