Researcher: Marios Savvides
Research Area: Security of Cyber-Physical Systems
While humans can often recognize the same person even when changes in appearance, such as facial hair or glasses occur, many face recognition algorithms fail in attempts to match a face because they do not compensate for the differences in appearance. We aim to add intelligence to the current systems by building robust classifiers that can determine these “soft biometric” traits. Soft biometric intelligence can be integrated to improve accuracy in facial matching where match detection in conventional systems may fail due to constantly changing soft biometrics. Additionally, soft biometrics can also be used in real-time to narrow identity searches of unknown individuals. When the only a description of the subject is available, e.g. “Caucasian male with glasses and a moustache,” the use of soft biometric identification can greatly and accurately assist in narrowing the search space.
Outcomes: a robust soft-biometrics system that allows classifying gender, ethnicity, age, beard, mustache, glasses, etc. from given face images.