Programmer Helps Solve Google Privacy Problem

Privacy is a problem because in some areas where the Street View cars operate, the pictures are so sharp that individual people can be seen and identified, or license plates read. To prevent this Google blurs the faces of people in the pictures.

Follow us

But the blurring isn't always perfect. For example, the red "stop" signs on school buses are sometimes blurred, even though they contain no identifying information. For people, often body shape and clothing are enough to identify someone.

What Flores hit on was using the multiple angles that the street view vehicles take pictures from as a base from which to reconstruct the background image. By looking at a wall, for example, from different angles, the software can figure out what it would look like without the pedestrian in the way.

Belongie said the results aren't perfect. "An artist using photoshop could make a more aesthetically pleasing result," he said. But at the end of the process the people are gone from the picture. It's an automated process, so occasionally one sees images of dogs with leashes that end mysteriously in the air. "We used to joke that it was like the rapture and everyone was gone, and dogs stayed because they didn't have souls," he said.

There are some drawbacks to the technique. Belongie noted that some people might want the images to be full of people - if one is selling real estate, for instance, the impression of a lively neighborhood can be important.

Then there is the problem for Google. Faces and license plates were relatively easy to blur automatically, but removing people entirely could mean a torrent of requests, Belongie said. That might be one reason Google hasn't done this already.

Belongie also noted that the kind of cameras Google uses on the street view vehicles tend to be higher quality in cities such as New York or San Francisco, so the problem of identifying pedestrians would tend to be focused on such large urban areas.

In an email, a Google spokesman asked how to contact the two researchers, but declined to comment on the technology itself. Google has no current plans for this project, she said.