Quasi-Invariant Matching for Robust Biometric Identification

DoD and the Army in particular are working to create effective biometric systems for positive identification of users for access to facilities and for the use of equipment including tactical gear, vehicles and weapons. Commercial off-the-shelf (COTS) biometrics capabilities are scaling to meet industrial needs such as employee identification, and many COTS products are available for fingerprint identification. However, despite recent advances, existing systems still lack sufficient accuracy and tolerance to handle the distortions expected under battlefield conditions. The battlefield gives rise to poor biometric signature acquisition and requires much shorter response times than are acceptable under commercial conditions. Innovative technological developments are required to ensure that a poorly read measurement of an authorized user's biometric signature is accepted, without giving access to unauthorized users including hostile forces and indigenous civilians.

In order to meet the requirements of biometrics under battlefield conditions, IET worked under the SBIR program to apply the theory of observational quasi-invariants and hierarchical Bayesian inference, which derive from advances in pattern recognition developed for computer vision applications, to the biometric challenge of correctly associating human identification with measurements taken under a wider range of distortions.