Researcher: Marios Savvides
Research Area: Security of Cyber-Physical Systems
Scope: Soft-biometrics aims to develop a robust and intelligent fusion method for multi-biometric identification using a set of less distinctive biometric features. Current face recognition systems do not consider changes in soft-biometric attributes (such as gender, ethnicity, age, facial hair, etc). While humans can often recognize the same person even with changes in appearance, the majority of the face recognition algorithms fail in such situations since they do not compensate for these differences in appearance. Integrating soft-biometrics into these systems to account for these facial variations will greatly enhance the performance of face recognition when applied to real-world images. We aim to achieve this by adding intelligence to the systems by building robust classifiers that determine these “soft-biometric” traits such as gender, ethnicity, age, facial hair and glasses.
This method is especially useful in law enforcement, where the identities of subjects are generally unknown and descriptions are based on these less distinctive soft-biometric attributes. Our system can narrow down identity searches in large databases by cross-referencing these attributes with corresponding images present in the database.
Outcome: The result of this project will be a soft-biometric identification system that will automatically extract attributes and use them to intelligently sort through large datasets and increase face recognition performance by adding Artificial Intelligence to current facial matchers.