Enhanced Learning of Sensor Fusion for Human Authentication
Abstract
User authentication is a critical component in supporting numerous military and civilian applications. The increasingly pervasive wireless networks make it even easier to conduct attacks (e.g., identity theft and forging digital commands in tactical fields) for new and rapidly evolving adversaries. Although password-based and fingerprint-based authentications can serve as the first line of defense, the fast-paced technology-oriented modern warfare has a pressing need for multi-modal user verification in every aspect of military applications to ensure smooth executions in tactical operations. Existing user authentication systems often need to employ expensive and specialized hardware. Also, many of them use only a single type of biometric traits. Under mercurial environmental conditions, they can suffer loss in effectiveness and become vulnerable to various kinds of identity-based spoofing attacks. With the rapid advancement of wireless technologies, there are various sensors embedded in mobile devices (e.g., smartphone) designed for specific mobile applications. These embedded sensors have the potential to work together to perform enhanced authentication through multi-modal user verification. This kind of authorization does not require dedicated infrastructure nor customized hardware. Therefore, it becomes extremely appealing especially in todayÕs technology-oriented warfares where military personnel carry various mobile devices around the clock. Toward this end, we propose a learning-based user authentication framework leveraging multi-modal user verification grounded on humanÕs physical traits extracted from existing mobile devices without requiring specialized hardware or verification infrastructure. Our sensor-fusion based machine learning techniques will add two important security aspects in user authentication: continuous user verification exploiting unique gait patterns and online signature verification via touch-screen sensing. Our research goal is to (1) exploit ensemble learning of information from multiple classifiers on single sensor data to combat the inherent problems coming from a single biometric indicator; (2) use confidence distribution combining to aggregate decisions from multiple sensors to produce more powerful and robust matching results to achieve enhanced user authentication. This proposal emphasizes the importance of integration of theoretical study and system approach by validating our proposed framework through software implementation and testbed development. And the opportunities of technology transition to ARO will be explored.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 14, 2019
- Source ID
- W911NF1810221
Entities
People
- Yingying Chen
Organizations
- Army Contracting Command
- Rutgers University
- United States Army