Dec. 9, 2008 -- Airport security guards could soon be pointing lasers at your head, if the latest research on face recognition holds up.
Security services around the world are devising software to match CCTV photos of faces with image archives, to catch known criminals and terrorists. However, the technology struggles to overcome problems caused by variations in lighting and the position of people's faces between photos.
Now US research suggests that 3D laser scans could be a more robust option in tricky real-world conditions.
Dirk Colbry, at Arizona State University, and colleague George Stockman at Michigan State University, took 300 laser scans of 111 different faces using a commercial scanner.
The device's stationary camera records a series of images while a horizontal plane of laser light passes over a person's face. Software then extracts a three-dimensional model of the face from the photos.
Colbry and Stockman showed that they could match different scans of the same face even when lighting was different, or the angle from which the images were taken was off by as much as 30 degrees.
Facial recognition accuracy is gauged by a figure called "equal error rate" - the point at which false positives equal false negatives. The 3D scans achieved a figure of 1%, which is considered good for facial recognition systems.
Software recognises faces by matching "anchor points", such as the tip of the nose or the corners of the eyes, in two photos or 3D scans. Anchor points in 2D images appear to be more strongly affected by differences in lighting or angles than in 3D images, Colbry says.
There are problems with the approach, however. Laser scans currently take 2 to 5 seconds, making the technique inadequate for recognising passing faces in a crowd.
The second problem is cost. Colbry says scanners are currently priced at about $50,000. That will need to fall to $5000 or less to be generally useful, he estimates. However, like many other technologies, scanners are becoming inexorably cheaper.
Colbry suggests that it is still worth preparing for the day that 3D scanners are fast and cheap, rather than perpetually struggling to improve matching between 2D photos.
William Smith, a lecturer on computer vision at the University of York, says the new work confirms that 3D images can be useful for verification. But their weakness is that subjects have to cooperate, whereas photos can easily be taken and compared covertly.
"If the subjects are aware of the imaging procedure and willing to take part, why not use more robust fingerprint or iris scans?"
However, Colbry thinks that scanners could eventually capture profiles without the subjects suspecting. Security checks often force people to stand still in queues, providing good opportunities for covert imaging, he says.