May 18, 2001

UCSC researcher focuses on clearing blurry pictures

By LOUISA DALTON
Sentinel correspondent

The video from the family picnic is back, and grandma’s so blurry she could be grandpa.

"Really, there’s no mechanism to improve that picture now," said Peyman Milanfar, a UC Santa Cruz electrical engineering professor. So Milanfar has just come up with a computer program to do just that: Take fuzzy throwaway pictures or videos and sharpen them into clear, quality images.

The program could be used not only to distinguish grandma from grandpa, but also to make out the license plate on a speeding car or sharpen the ultrasound image of a baby.

One of the greatest advantages to the technology, Milanfar said, is that you could use cheap cameras to take great pictures. For example, the military may not want to send expensive camera equipment into a sensitive area to monitor road movement. But the military does want to know from that camera if the writing on the side of a passing truck says United Nations or not.

The trick to Milanfar’s program is that he uses two or more poor pictures to make one good one. If you take a few pictures in a row, which is exactly what a video does, there always will be a slight movement of either the camera or the object or both. Milanfar’s program uses that slight shift to better define the object.

"We’re giving up resolution over time to get better resolution in space," Milanfar said.

Video cameras now usually have a feature called vibration or motion control, which corrects for small, unintentional jittery motions. Most people don’t want the shaky, unprofessional look.

But jittery video is actually a great thing for enhancing resolution later, Milanfar said.

"We want tiny movements in all directions," he said.

The combination of many slight movements makes for a well-defined still shot, or a high-resolution video with fewer frames.

The original picture doesn’t need to be video, said Milanfar, who collaborated with Nhat Nguyen at KLA-Tencor Corp. in Milpitas and professor Gene Golub at Stanford University. It also can be two still shots taken in a row.

The picture-improving technology is "still at a research stage," Milanfar said, and may not be generally available for another five to 10 years. However, his computer program has made its use more feasible than ever. The major step forward, said Milanfar, who published the breakthrough in the April issue of an engineering journal, is that they can perform the computations in a reasonable amount of time. The picture of the future looks clearer than it ever has.

 

Contact Louisa Dalton at ldalton@santa-cruz.com.