Can Artificial Intelligence Become a Tool to Help Detect Art Forgeries?

Learned art experts have long used the study of brushstrokes to help authenticate questionable paintings by the great masters. But acquiring a sufficient level of expertise took many years, if not an entire career. In the end, the few such sophisticated art scholars were only found in prominent art institutions and universities.

That’s all about to change.

Quite soon, almost anyone will be able to instantaneously access art knowledge and experience that previously took a lifetime to acquire.

On February 2 in New Orleans, the 32nd AAAI Conference on Artificial Intelligence will hear from Rutgers University Professor of Computer Science, Ahmed Elgammal, head of the Art and Artificial Intelligence Laboratory.

The technical solutions in his paper, “Picasso, Matisse, or a Fake? Automated Analysis of Drawings at the Stroke Level for Attribution and Authentication,” promises powerful new capabilities to better detect fraudulent artworks. If further developed, this new anti-fraud technology will complement other lab-based methods, providing stylistic analysis that currently can’t be done. In addition, it will be cost-efficient.

The ramifications are enormous for the art world, where major fraud scandals are an increasing occurrence, and where experts have been sounding the alarm that the market is inundated with fakes, themselves often made possible thanks to new technologies.

Through his company, Artrendex, Dr. Elgammal teamed up with the Atelier for Restoration & Research of Paintings in the Netherlands to develop an AI system that analyzes individual strokes in drawings, comparing them to a large number of strokes by different artists using machine learning.

According to Dr. Elgammal’s paper: “The goal is to test the hypothesis that artists can be identified based on individual strokes, which remains largely untested scientifically. The goal also is to build a robust AI system that can help in attribution and authentication of artworks, mainly based on the characteristics of strokes, and complementing other existing technologies.”

Scholars utilized 300 digitized drawings that had over 80,000 strokes. Mainly consisting of Picasso, Matisse and Schiele drawings, “the experiments show that the proposed methodology can classify individual strokes with 70%-90% accuracy.”

For the first time, technology can attempt to replace human art experts at the visual level.

While Dr. Elgammal’s technology will complement existing anti-fraud analysis techniques, it promises to be the first major step toward helping detecting a fake automatically, and should catch most quite easily in the early stage.