The research team, composed of people from ING, Microsoft, Technical University Delft, Amsterdam's museum Het Rembrandthuis, and the Mauritshuis art museum in The Hague in the Netherlands, studied Rembrandt's work and used computer programs and a 3D printer to mimic the artist's style.
The researchers started by collecting data pixel by pixel from all 346 Rembrandt paintings with 3D scanners and a deep learning algorithm. Once the database was complete, the lead developers, Morris Franken and Ben Haanstra, started to identify the finer points of the painter's work. Using statistical analysis and a facial recognition algorithm, the developers went on to "extract the features that make Rembrandt Rembrandt," Haanstra said during a short film about the process.
After establishing Rembrandt's unique talent for depicting human emotion, the team used the data to determine the subject of the new painting. The data showed that Rembrandt usually painted Caucasian men with facial hair, 30 to 40 years old, wearing dark clothing with a collar and a hat, facing to the right.
The researchers used a height map to mimic Rembrandt's brush strokes. The next step was to upload the data to a 3D printer, which was filled with special UV ink, and watch as the printer laid down 13 layers of ink. The final painting is composed of 148 million pixels.
After 500 hours of work over 18 months, the portrait, titled "The Next Rembrandt," was unveiled in Rijksmuseum in Amsterdam this week.
Watch the video below.