AI system can spot a fake drawing from the authentic with a single stroke

The method of studying drawings is inspired by pictology.

The 1954 Pablo Picasso painting "Femme accroupie (Jacqueline)", a portrait of Jaqueline Roque, Picasso's final great muse and eventually his second wife, at Christie’s London. (AP File Photo)

A new research says an artificial intelligence (AI) system can detect forgeries of drawings with a single brush stroke.

The AI studied 297 digitised sketches and drawings, and 80,000 brush strokes by artists Pablo Picasso, Henri Matisee, Amedeo Modigliani and Egon Schiele to spot a fake with 100% accuracy.

Researchers from Rutgers University and the Atelier for Restoration & Research of Paintings in the Netherlands trained algorithms to study the drawings, differentiate characteristics of brush strokes, and then look at traits unique to a specific artist. The combination yielded absolute results.

The method of reading into drawings is inspired by Maurits Michel Van Dantzig's study of pictology. Dantzig proposed looking at strokes made by an artist and studying features of art carefully to recognise an artist's style of drawing and composition. For instance, renaissance painter Leonardo da Vinci often painted his signature curls that experts say were his favourite.

Contrary to pictology in which a human studies paintings, researchers say the AI system can give a stylistic analysis of art in a scientific manner, and it may cost less than other technical analysis methods.

"A human cannot do that," Ahmed Elgammal, a professor at Rutgers University and one of the paper's authors, told MIT Technology Review.

However, the paper used a sample of drawings and sketches that could be very different from studying paintings in which artists merge brush strokes, a style employed by impressionists.