New Software Lets Users Insert Virtual 3-D Objects Into Photos
|
|
CREDIT: Kevin Karsch |
The room looks normal enough, except for the shoulder-height dragon statue in the middle, between the couch and the dining room chairs. Weirder yet? While the room is real, the statue is not. It's a 3-D drawing that researchers at the University of Illinois at Urbana-Champaign have inserted into the picture.
The researchers have created a simple computer program that lets users add 3-D drawings of objects into photos. The program can even incorporate animated objects, such as a ball rolling across the floor. The ball will move behind walls and cast shadows realistically no photo-editing skills needed. The added dragon statue casts a shadow on the wall in front of it and its rump reflects the light from the lamp behind it.
The work is excellent, Hany Farid, who wasn't involved in the research, told InnovationNewsDaily in an email. Farid is a Dartmouth College computer scientist who studies computer-generated images.
The researchers who created the program will present their work in December at an Association for Computer Machinery conference in Hong Kong. They hope the program will be used in videos, computer games and a home-decorating app that people could use to add furniture to a photo of any empty room. The team just began talks with the University of Illinois about selling their technology to a commercial company, said Kevin Karsch, a doctoral student in computer science who led the research. He doesn't know how long it will take to see his program on store shelves.
Of course, moviemakers often add computer-generated imagery (CGI) to real scenes. The professional programs they use require time and expertise, however, and the user needs to upload several pictures and measurements of the actual room.
The Illinois program requires just one photo and the user never has to set foot in the actual room. By clicking on the picture, users specify where the edges of the room are, the shape of light sources and the shape and height of surfaces on which they want to place things. That's enough for the computer to light the scene and bounce balls off of sofas in the room. You can make these scenes in anywhere between one and five minutes, Karsch said.
The results are not as detailed as professional CGI, but they're nevertheless convincing. Karsch and his colleagues tested just how convincing they are by asking 30 volunteers to pick which of two pictures was the real one, in a series of photos where one photo was real and one was a real-synthetic hybrid.
Overall, people choose Karsch's hybrids as real 20 percent of the time. In photos with more than two added objects (a wine glass, a flowerpot, a pink ball), people chose the hybrid photos 28 percent of the time. In those photos, Karsch's method outperformed a standard lighting algorithm called light probe, which people mistook for reality 25 percent of the time.
The work, which went viral online after Karsch posted about it on the video-sharing website Vimeo, has drawn some skeptical questions. What hath God wrought? wrote The Atlantic online. Who would ever believe any video evidence now? asked a commenter on the technology forum Slashdot.
But don't expect 3-D politicians to be inserted into compromising scenes, said Cynthia Baron, who directs the digital media program at Northeastern University. She wrote a book about photo editing fraud, "Adobe Photoshop Forensics: Sleuths, Truths and Fauxtography," in 2007. If the program is sold commercially, she expects the response to be mostly benign. I expect to see much more realistic ghosts around Halloween ... and a lot more faked images on the Internet, she told InnovationNewsDaily.
The major limitation of Karsch's program is that it needs fully rendered 3-D images to insert, Baron said. At this point, only expert digital artists can make convincing 3-D images. Realistic 3-D people are beyond anybody's abilities right now. Moreover, such renderings are stored in virtual libraries, so should anyone need to prove a photo were altered, they could find the reference object in a library.
Baron expects mostly fun, creative things from the program and speculated that it might make people more careful and thoughtful when they encounter a picture.
But that's nothing new, Karsch, the software's creator, said. People have been doubting pictures since photography has been developed.
Karsch said his team is working on enabling synthetic object insertion in video, where both the background and the inserted objects can move. Check Karsch's Vimeo account for updates. He expects to post results in the next week.
