- 11:17 21 August 2008
- NewScientist.com news service
- Colin Barras
Digital photographers are used to tinkering with their photos at home. Now, amateur video makers can create movie special effects thanks to prototype software from Microsoft.
Their application, called Unwrap Mosaics, makes adding a moustache to a family member in a video as easy as scrawling on a digital picture. A video (right) shows it in action.
For example, drawing a tattoo on a person’s arm requires editing only a single image. The programme makes the change throughout the video, making it ripple realistically with their skin.
A video is simply a sequence of still images that capture the movement of objects over time. But editing the appearance of objects in even a short movie is difficult because every frame shows objects differently and must be edited separately.
“We thought it would be good if you only had to draw your edit once and [the software] propagate the information from that single edit through the whole video,” says Alex Rav-Acha at the Weizmann Institute of Science in Rehovot, Israel.
Rav-Acha and colleagues at Microsoft Research in Cambridge, UK, have done just that. Editing objects using their software is a one-step process.
Unwrap Mosaics was developed by Rav-Acha, Andrew Fitzgibbon, Pushmeet Kohli and Carsten Rother. It virtually strips the skin from a selected object in a video, producing a 2D surface that can be easily edited using photo-editing software.
Rav-Acha likens that surface to a tiger-skin rug: a flat surface that once covered a 3D object. The software reverses the process after any edits are complete, wrapping the edited skin back around the 3D object. Because the edits are attached to the skin, they then move with the object throughout the movie.
The 2D skin has to capture every inch of the 3D surface for edits to match the object when it is wrapped up again. For example, an “unwrapped” face must include information from both sides of the nose so that changes to the nose remain realistic even when the head turns from side to side.
To fully unwrap an object the software tracks 5000 points spread evenly across its surface, making sure that all of those points are present in the final 2D skin.
That means the final unwrapped skin contains more information than any individual video frame. For instance, if the video shows a person’s head from the front and then turning to the left, the unwrapped skin contains a full view of the ear as well as a forward view of the eyes, nose and mouth.
The user makes their edits to the skin before the unwrapping process is reversed to send those changes throughout the final movie.
Rav-Acha and his co-workers have successfully tested their software on a range of videos. “There are professional tools that can do these kind of edits but it’s an incredibly skilled procedure,” says Fitzgibbon. “We believe that with this procedure anyone should be able to edit video.”
Alex Rav-Acha’s team presented their work at the SIGGRAPH conference in Los Angeles last week.