Multimodal Biometric Watermarking-based Transfer Learning Authentication

Abstract

Though biometric-based authentication systems have inherent advantages over conventional authentication systems, which use passwords and ID cards, these systems cannot ensure the security and privacy of biometric data stored in their databases. Recently, several watermarking-based algorithms have been efficiently used to protect biometric templates. However, these methods also require storing the watermarked biometrics for the matching purpose of the query during the authentication phase. The paper aims to develop an approach for authenticating watermarked biometrics using transfer learning without the need to store biometric data. Benchmark face and fingerprint databases are employed to conduct the experimentation. The obtained results validated the proposed approach's ability to discriminate between different users with a performance accuracy rate achieving 99.17% while protecting the user's biometrics.