A computer scientist at the University of Minnesota in Minneapolis-St. Paul is using techniques derived from fluid dynamics to re-create, or "inpaint," the missing parts of deteriorated images. Current methods of digital correction for old movies, paintings, and photos require laboriously matching colors, shadows, and textures. The new approach can do most of the job in seconds.
Guillermo Sapiro and his colleagues think of the damaged areas of a photo or painting as holes. "The information around the hole is like water that is held back by a wall. I want the information to flow into the hole so that it continues the patterns from the surroundings," Sapiro says. That stream of data re-creates what once might have filled the empty space. Shapes and contours are continued; colors are matched and blended. Sapiro's software can fix scratches in photos, sharpen digital images—even remove objects from a scene. "There are always things that moviemakers didn't want to be there—say, an actor leaves his shoes in a scene," he says. His technique might also prove useful for validating sensitive visual data, such as evidence from a security camera. "Inpainting puts artificial information into an image and makes it look normal, but it is still a human-made process," he says. Running the process backward could reveal whether an image has been altered.
Defaced crocuses (top) look good as new after data cleanup (bottom).Photograph courtesy of Guillermo Sapiro/University of Minnesota.
