3D Face Shots Snapped by Single Camera

Feb. 6, 2007 — An imaging technology originally used to find bumps and flaws in industrial surfaces is now being turned on the tiny details of the human face to produce three-dimensional pictures.

Such detailed maps of a person's features could be used for security purposes in places like airports and banks, but it could also lead to better imaging and diagnosing of skin trauma such as burns.

The images can be combined with software that adjusts for light or other conditions if security officials need to compare a photo with one taken previously with another camera or under different conditions.

"You can create how that person would have looked if seen under different viewing conditions and then match it against a database of suspects...or you may use computer graphics to see how the person looks with glasses or moustache," said Maria Petrou, team leader on the project at Imperial College London.

The technique, called photometric stereo, uses a fixed digital camera and at least three lights placed around it to illuminate the face from different angles.

The lights are synchronized to flash very quickly in succession, in a few hundredths of a second, so the person being photographed only perceives one flash. But the computer picks up digital data for all lighting angles.

A program written by Petrou and her colleagues analyzes the shadows and highlights, and then combines them into one three-dimensional image.

Other recognition systems are capable of producing three-dimensional images, but most rely on two or more cameras.

And because each camera snaps the same photo from a different location, the resulting images contain slight disparities in the location and color of pixels that correspond to the same physical point on the person's face.

Identifying the pixels in each image that correspond to each other is difficult, but necessary to create a three-dimensional image.

Photometric stereo overcomes the pixel-matching problem because one camera takes multiple shots.

"Potentially, it could produce an excellent and cheap...system that would be very tough to beat simply because of the vast amounts of 3D detail it captures. It's also non-contact and quick," said Michael Chantler of the School of Mathematical and Computer Sciences at Heriot-Watt University in Edinburgh, Scotland.

The challenges, he said, will be in quickly comparing the captured image to others in a database as well as coping with variable facial expressions.

Petrou and her team hope to have a working prototype in three years.