How new AI research is changing the way curators and collectors view attribution



“Accuracy when using AI and neural networks relies to a large extent on the quality of the data,” she said. “We use a two-step authentication method to collect reliable data that filters out spurious and unreliable images, ensuring that training images are referenced to the existing catalog raisonné. We work with art historians to provide our clients with an unbiased report on an individual painting or work.

She noted that their method can provide collectors with precise results that can indicate both an original work of art or an artist’s style, provided there are enough images to train the neural network. According to Popovici, around 100 images are needed to get accurate results, but they often work with a lot more. For their database on Cézanne, for example, they work with 850 high-resolution images of the artist’s work.

In the past, the recognition of art has been accredited with authenticate a disputed painting by Peter Paul Reubens, but admits the process isn’t always perfect. “AI has its limits when it comes to authenticating works of art,” Popovici said. “But we think it may offer just one method of providing insight into who created a work of art.”

A heat map of the same Van Gogh self-portrait. Photo: Artistic recognition.

Singer agrees, noting that while the method may not be foolproof, AI and machine learning have come a long way in artistic authentication.

“We like to think of it as just one tool among many,” Singer said. “The goal is to refine the method and improve it over time. To that end, he and his team worked with Madrid-based Factum Arte to transform their custom Lucida 3D scanner into El Greco’s. Portrait of Juan Pardo de Tavera (1609), which was damaged and restored after the Spanish Civil War. The painting was photographed before and after restoration, which makes it easier for researchers to identify areas of the canvas not painted by the artist’s hand.

“In this case, we can look at El Greco from a conservator’s perspective and help him identify areas unpainted by the artist,” Singer said.

This project has led to a new study that Singer and his team are working on, also with students at the Cleveland Institute of Art, who they hope will be able to identify the different hands at stake on a single piece of art. .

During and after the Renaissance, Western artists often employed assistant schools in their workshops to render various areas of a canvas. All of those hands, according to Singer, can now get the recognition and award they deserve.


Previous Burst pipe damages computers donated to Calgary students
Next 10 Minute Genius: Toby Walsh on mutant algorithms