Siggraph: Microsoft the new research powerhouse in graphics?

Computerworld - Think of companies doing state-of-the-art research in computer graphics, and you might guess names such as Industrial Light & Magic, Adobe Systems Inc., or Nvidia Corp.

But at this week's Siggraph 2007, for more than three decades the most prestigious international conference for academic and industrial computer graphics research, Microsoft Corp. researchers will be presenting or co-presenting one out of every eight papers.

That is "more papers than MIT or Stanford or any other institution that does research," said Richard Szeliski, head of the interactive visual media group at Microsoft Research. "What we produce is of the same quality and caliber as the best universities in the world."

At last year's Siggraph, Microsoft researchers presented or co-presented an even more impressive one out of five papers. They included one co-authored by Szeliski called "Photo Tourism: Exploring Photo Collections in 3D" (download PDF), which used the same PhotoSynth technology that NASA said yesterday underlies 3-D views of the space shuttle Endeavour.

To underscore that this year is no fluke, Microsoft also said today that Weta Digital Ltd. the visual effects company co-owned by Lord of the Rings and King Kong director Peter Jackson, will license graphics technology developed at Microsoft Research.

Weta, whose special effects for the Rings trilogy won multiple Academy Awards, will use the technology to help create graphics and special effects for movie, commercial and other multimedia productions.

PhotoSynth is image stitching technology taken to a logical extreme. It is a pet area of Szeliski, who joined Microsoft from Digital Equipment Corp. in 1995 and has seen the technology progress from being unable to stitch together two pictures taken at different times of the day to handling 3-D panaromas.

At this year's conference in San Diego, Microsoft researchers are showing off what is arguably even more impressive research.

It includes image deblurring software (to learn more, watch this Windows movie presentation) that combines a pair of pictures taken in low light -- one blurred from subtle camera shake during a long exposure, the other very grainy because it was taken at regular speed -- to reconstruct a single, high-quality image.

Essentially, the software figures out how the camera jiggled and then corrects for that. It then takes the shot, now cleaned up and boasting fine detail, and overlays it on top of the stable but grainy image.

"The results are stunning," Szeliski said. The software, developed by Microsoft and academic researchers in Beijing and Hong Kong, can improve images as coarse as those taken by today's 1-megapixel cell phone cameras, he said.

In several years, we might be able to buy digital cameras that, in low-light conditions, automatically take one blurry and one grainy photo and merge them together, Szeliski said.