AI Restore Damaged Art

Photo credit: Daily Mail
Researchers at MIT’s CSAIL have unveiled “RePaint,” a system that can reproduce damaged artwork using an artificial intelligence-guided 3D printer. This means color-accurate reproductions, even in less than optimal conditions, such as low-light. “If you just reproduce the color of a painting as it looks in the gallery, it might look different in your home. “Our system works under any lighting condition, which shows a far greater color reproduction capability than almost any other previous work,” said Changil Kim, one of the MIT researchers that published a paper on the system. Unlike other systems, RePaint works by stacking ten different transparent inks in thin layers, and then the AI predicts the ideal stack needed to generate the desired colors. Read more for additional examples and information.


AI Restore Damaged Art
AI Restore Damaged Art
MIT researchers used a special technique called “color-contoning’, and claim that RePaint is more than four times more accurate than state-of-the-art physical models at creating the exact color shades. The physical reproductions are business card-sized to minimize costs for the time being, or at least until they can get costs down.

AI Restore Damaged Art
AI Restore Damaged Art
“Color-Contoning” was combined with an old technique called “halftoning,” where an image is made by using little ink dots, rather than continuous tones, to better captured the nuances of the colors. A deep learning model was used to predict the optimal stack of different inks before being fed images of paintings.