Viisage, Siemens turn 2-D into 3-D
- By Dibya Sarkar
- Aug 12, 2004
Viisage Inc. and Siemens AG officials are jointly developing 3-D facial recognition mapping that could further enhance the accuracy and performance of the biometric technology.
Rather than take true 3-D images of people, Viisage uses a process called 3-D warping, in which algorithms transform 2-D pictures into images that behave similar to 3-D composites. But since such composites are never 100 percent complete, Viisage officials are using Siemens technology "to fill in the blanks with real data, not guesswork" to enhance the quality of the composites, said Mohamed Lazzouni, Viisage's chief technology officer and vice president of engineering.
"That's really the big problem we've solved," he said.
The only alternative to existing 2-D pictures would be new images of individuals who are already registered in existing databases, Lazzouni said.
"Instead of having databases of millions and millions of people that have digital images that are 2-D, you have to re-enroll every human being with a 3-D image," he said. "The question is who can do that? The answer is nobody."
Many police departments, corrections facilities and other organizations across the country use facial recognition technology for authentication and verification. The technology measures certain physical characteristics of the face, but illumination, facial expression and pose variation can disrupt the ability to match well.
Lazzouni said the 3-D technology model would basically address the pose variation problem, which is "when people tilt their heads from left to right, top down or take an angle, that is a little bit severe."
However, using the technology in the real world is still more than two years away, he said. Governments, companies and other organizations might one day use 3-D sensors or surveillance cameras in highly secure locations to capture images of people. "The moment that takes place then we automatically have a position of advantage because now that 3-D captured image can be compared to a true 3-D representation of a good quality composite," Lazzouni said. "People who turn their faces as much as 45 or 50 degrees in one direction or the other could be identified successfully."
The two companies also are jointly developing Siemens' proprietary core technology for a 3-D sensor and will co-market it.
There are other facial recognition companies working exclusively on 3-D mapping, but they are mostly experimental and using small amounts of data, Lazzouni said. Viisage has deployments across the country and databases with millions of images that can be converted into 3-D images.
But Lazzouni warned that the 3-D technology is "no more than an incremental enhancement" upon something that already exists.
"It's going to work for specific cases if you meet the requirements well enough," he said. "You can't just simply throw it out there in the hope that it will enhance every case that you come across."