Carnegie Mellon to Present Realistic Photo Clip Art
Carnegie Mellon University computer graphics researchers have developed a way to easily edit photos using the millions of photos available on the Web, announced the university yesterday.
Products are chosen independently by our editors. Purchases made through our links may earn us a commission.
July 12, 2007 – Carnegie Mellon University computer graphics researchers have developed a way to easily edit photos using the millions of photos available on the Web. The university yesterday announced the Photo Clip Art and Scene Completion algorithms, which pull photos from Web databases such as Flickr to manipulate digital photos or prints, requiring no technical photo editing skills.
'Why Photoshop if you can 'photoswap' instead?' said Assistant Professor Alexei Efros of the Carnegie Mellon School of Computer Science in a CMU news brief yesterday.
CMU developed two systems: the first to add realistic objects to existing digital photos, and the second to fill in holes in damaged prints or files for a seamless composite photo.
Photo Clip Art can add object cut-outs to a digital picture. The system pulls photos from an image-based object library called LabelMe, a Web database full of labeled photos. The algorithm automatically resizes, blends, and adjusts scene lighting to make one seamless, realistic composite photo.
The other system, Scene Completion, uses an algorithm to patch up holes in prints or digital images. Scene Completion uses photos from the online sharing site Flickr as a database. The system, which doesn't require user intervention, then fills in missing sections by matching color and texture.
The Carnegie Mellon research team will present the abstracts "Scene Completion using Millions of Photographs" and "Photo Clip Art" next month at Siggraph 2007, one of the world’s largest computer graphics conferences.
Images courtesy of Carnegie Mellon.