Intelligence Value
The Odin program is developing biometric presentation attack detection technologies to help the government better detect when someone is attempting to hide their identity by spoofing biometric access control or authentication systems.
Summary
Biometrics are used by the U.S. government to verify identities and locate persons of interest. However, biometric presentation attacks, known as PAs or spoofs, can prevent correct identification. These PAs can also lead to unauthorized access to sensitive facilities, or information. Compromised biometric systems erode the trust of users. Odin’s goal is to identify known and unknown PAs in face, iris, and fingerprint biometric collection systems. Typical PAs utilize a prosthetic to conceal the subject’s biometric trait, or present an alternative biometric signature. Odin detects these PAs, with both software and hardware solutions, by applying deep learning and computer vision to visible or multispectral images. The results are used to discriminate between PAs and legitimate samples, based on prior knowledge of the attributes of a true sample, and normalcy modelling for anomaly detection.
Odin’s presentation attack detection system’s performance is quantified by the detection accuracy with a corresponding false alarm rate. This metric reflects the important balance for real-world usability. The final target metric is 97 percent average detection accuracy, at 0.2 percent false alarm rate. The unique methods developed and tested on large human-scale subjects under the Odin program have met or exceeded research goals and are positioned for deployment in real-world biometrics applications.
