USC experimental biometric suites go deep and broad to thwart fraudsters

Biometrics researchers at the University of Southern California boast 100 percent success at spotting attempts to cheat a facial recognition system they designed. Third-party testing by Johns Hopkins University backed up the results.

USC also received confirmation that its Biometric Authentication with Timeless Learner (BATL) project achieved a 99.36 percent iris recognition score and a 99.08 percent fingerprint recognition score.

The hardware and software project is the product of the Vision, Image, Speech and Text Analysis group, which develops spoof-resistant biometric authentication systems. It is part of USC’s Information Science Institute.

University officials have been discussing the project since at least February 2019. Their goal has been to integrate multiple kinds of scans to measure multiple biometrics to uncover spoofs that might fool less-sophisticated interrogations.

An example of a simpler technique commonly used now is employing an off-the-shelf digital SLR camera to record a face, even though such cameras can be fooled with pictures of a face and contact lenses.

Instead, iris and face sensor suites capable of analyzing multispectral data was built involving LED lights shining long-wave IR, near IR and visible light among other wavelengths. It has not been disclosed how long the system takes to recognize a face.

The sensor components measure “conventional and unconventional” biometrics, according to USC. They analyzed light bouncing off and heat energy radiation from skin, for example. Recording a spectrum of evidence would force fraudsters to imitate the same spectrum, making an attack more time and resource consuming.

Algorithms were written to make use of the new streams of data. The machine learning models “are evaluated in many different ways to make sure our final models are resilient” to different attacks, according to USC.

The no-contact fingerprint sensor suite measures more than static print patterns. It also notices blood motion, skin translucence and vein visualization.

Johns Hopkins’ Applied Physics Lab tested the setups for three weeks using data from about 700 participants.

Article Topics

authentication  |  biometrics  |  biometrics research  |  facial recognition  |  fingerprint biometrics  |  fraud prevention  |  iris recognition  |  spoof detection  |  University of Southern California