04. August 2020, 09:53 Uhr | Ralf Higgelke
With Beam Profile Analysis from trinamiX, a smartphone can be unlocked safely and reliably using face recognition even with a mouth-and-nose protection mask.
During the Covid-19 pandemic, most people are wearing a mask over mouth and nose. But is a user then able to unlock his smartphone by facial recognition? BASF subsidiary trinamiX has developed a special technique to overcome this issue.
Standard 2D facial recognition algorithms can be adapted to users wearing a protective mask over their mouth and nose. However, the security of those systems is then lower, as fewer landmarks (unique features of the user’s face) are recognizable for the system – the same is true for 3D authentication. As a consequence, a successful spoof attack is much more likely. Users could make use of alternative authentication technologies, but this could affect the user experience – and maybe also security.
This is where the algorithms of trinamiX come into play. Its Beam Profile Analysis technology correctly identifies whether a real human person is present during the authentication process. This technology is an active measuring principle, which means that the face to be measured is illuminated with a light source emitting invisible near-infrared light in a regular dot pattern. The reflection of each light spot is captured by the standard CMOS camera in the device and its beam profile is then analyzed.
Living skin has a unique backscatter pattern under near-infrared light that is independent of pigmentation or visible lighting. When combined with the existing standard recognition software of mobile and desktop devices, spoofing (the technical term for fooling the device) the system with a realistic full-face mask, 3D sculpture or even a detailed 2D printout becomes virtually impossible. As each invisible light spot provides an independent skin check, even a partially-covered face can be flawlessly identified as real or fake.
Beam Profile Analysis can be integrated seamlessly with standard 2D facial recognition algorithms.