Security Interview
Face Recognition Systems: Almost Ready For Prime Time
Interview with Christoph von der Malsburg
Prof. Christoph von der Malsburg, 60, heads the Institute for Neuroinformatics at the Ruhr University in Bochum, Germany and the Laboratory for Computational and Biological Vision at the University of Southern California at Los Angeles. He developed one of the world's leading face recognition systems and set up ZN Vision Technologies to market it. His process is now used for access control at airports in Berlin and by police forces throughout Europe
How does Phantomas, your automatic face recognition system, assist in police investigations?
von der Malsburg: Phantomas automatically sorts pictures of wanted people in a database according to similarities, and then displays them on a monitor for a viewer. The recognition software examines about 1,700 points on a face at a speed of up to 10,000 data records per second. This process has allowed police in the city of Dortmund, Germany to triple their recognition rate.
Can the system be used to detect criminals by using cameras at airports or other public places?
Malsburg: No. That wouldn't be possible at the moment because the processes that are in use today are still too rigid. In other words, they are not able to cope with people turning toward the camera or any changes in lighting. We need to achieve another technological leap before they'll be able to manage that.
What sort of technological leap?
Malsburg: Photos extracted from video are by nature different from a picture in a database. To overcome this obstacle and be able to recognize a face, it must be possible to display the photo in 3D form. What's more, the light in the photo must also be changed to make it identical to the lighting in the picture stored in the database. I think we will reach this point in a year or two.
How will things develop after that?
Malsburg: Automatic recognition will not be limited to facial features; it will include other factors, just as the human brain does in real life. Humans don't just take in faces when they recognize a person; they also notice things like body size, clothing, movement or gestures. Developments would seem to indicate that these factors will be included in biometric processes in the future.
"Phantomas" can automatically compare 10,000 faces per second
Biometric processes are expected to be convenient for the user, guarantee a high level of security, and be inexpensive. Is all that possible?
Malsburg: Currently these three requirements cannot be met simultaneously. On the technical side, I don't see any great difficulties in making biometric processes comfortable, convenient and secure. However, the costs for some processesincluding face recognitionwill have to fall considerably. But I think that's possible. In my opinion, it depends on whether or not cameras with integrated computing units are mass produced. As soon as about a million are sold, the price per unit could fall to about 10 . That would open the floodgates for use of face recognition systems in all sorts of applications.
Interview by Rolf Sterbak