Genface: Improving Cyber Security Using Realistic Synthetic Face Generation

Recent advances in face recognition technology render face-based authentication very attractive due to the high accuracy and ease of use. However, the increased use of biometrics (such as faces) triggered a lot of research on the protection biometric data in the fields of computer security and cryptography.Unfortunately, most of the face-based systems, and most notably the privacy-preserving mechanisms, are evaluated on small data sets or assume ideal distributions of the faces (that could differ significantly from the real data). At the same time, acquiring large biometric data sets for evaluation purposes is time consuming, expensive, and complicated due to legal/ethical considerations related to the privacy of the test subjects. In this work, we present GENFACE, the first publicly available system for generating synthetic facial images. GENFACE can generate sets of large number of facial images, solving the aforementioned problem. Such sets can be used for testing and evaluating face-based authentication systems. Such test sets can also be used in balancing the ROC curves of such systems with the error correction codes used in authentication systems employing secure sketch or fuzzy extractors. Another application is the use of these test sets in the evaluation of privacy-preserving biometric protocols such as GSHADE, which can now enjoy a large number of synthetic examples which follow a real-life distribution of biometrics. As a case study, we show how to use GENFACE in evaluating SecureFace, a face-based authentication system that offers end-to-end authentication and privacy.