Video-based face verification for biometrics

Secure and accurate determination of user identity is central to the financial services industry. Traditional approaches to authentication include passwords, PINs, or token-based systems. With the ever increasing risk of fraud, and the consequences of security and privacy breaches, it is important that traditional authentication systems are enhanced with biometric modalities. They offer additional and more convenient means of authentication. Face biometric authentication is uniquely positioned for deployment in financial services environments as it can operate in both unconstrained and constrained environments. Examples of unconstrained environments in banking are identification from surveillance cameras and remote client authentication using cameras on mobile devices. Most current face recognition systems do not work well in uncontrolled environments, or they require a large amount of training data before they can be deployed.

The goal of this project is to explore recent advances in computer vision for robust face biometrics verification from video. Deep learning will be leveraged to achieve high accuracy in unconstrained environment. Innovation of this project potentially includes efficient approach for deep learning and adversarial robustness of deep learning for face biometrics. This project can potentially create innovative technologies for user authentication. Such technology could have both scientific value for the computer vision, machine learning and biometrics research community, as well as practical benefits for banking services.