A computer-generated portrait using machine learning algorithms created by The Next Rembrandt project has been shown by a team from Delft University of Technology in the Netherlands. The computer analyzed the work of the Dutch master Rembrandt, thoroughly tagged by humans.
“We researched the entire collection of Rembrandt’s work and studied the content of his paintings pixel by pixel,” the project team explains. “To get this data, we analyzed a wide variety of materials, such as high-resolution 3D scans and digital files, that were scaled up by deep learning algorithms to maximize resolution and quality.”
Because much of Rembrandt’s work was portraits, the team of researchers – supported by Microsoft, financial company ING and the Dutch museums Mauritshuis and Rembrandthuis – decided to focus on the portraits of the great man for analysis and recreation.
After considering age, demographics and other characteristics of the portraits, the researchers decided what they wanted the computer to paint: “a portrait of a white male with facial hair, between 30 and 40 years old, wearing black clothing with a white collar and a hat, looking to the right.”
To make the resulting photo more closely resemble the painter’s style, the researchers scanned the surface texture of his existing masterpieces to mimic the layers of paint.
“We created a height map using two different algorithms that found texture patterns of canvas surfaces and paint layers,” the team explains on the project’s website. “That information was converted into elevation data, which allowed us to mimic Rembrandt’s brush strokes.”
The resulting 3D-printed painting will be publicly exhibited and more details will follow soon, according to The Next Rembrandt team.