(from ACM TechNews, July 11 2007)
Carnegie Mellon University computer graphics researchers have developed systems for editing and altering photographs that automatically find images that fit with the original photo. The systems create well-blended images with minimal user skills, unlike traditional photo editing that can require a significant amount of skill. "We are able to leverage the huge amounts of visual information available on the Internet to find images that make the best fit," says assistant professor of computer science and robotics Alexei A. Efros. "It's not applicable for all photo editing, such as when an image of a specific object or person is added to a photo. But it's good enough in many cases." One system, called Photo Clip Art, uses images from the Web site LabelMe, which has thousands of labeled images, to add images to photos. For example, a picture of an empty street may be filled with images of people, vehicles, and even parking meters. To make the resulting picture as realistic as possible, the image analyzes the original photo to determine camera angle and lighting conditions. Then the system looks in the clip art library for appropriate images that fit the dimensions. The other system, called Scene Completion, uses millions of photos from the Web site Flickr to fill in holes in photos. Frequently, photo editors try to fill in the hole with a section from the same picture, but Efros says a better match can be found in a different photo. Efros and his colleagues will present papers on the two systems at the ACM SIGGRAPH annual conference, Aug. 5-9 in San Diego.