Face recognition is one of the foremost applications in computer vision, but faces are inherently sensitive signals that comprise personal identifiable information. Therefore, they must be protected not only when they are stored and transferred, but also while they are processed in an untrustworthy outsourced environment. This is one of the goals of Secure Signal Processing (SSP). This talk presents the privacy problems of outsourced biometric verification, with a focus on face biometrics, and the different alternative approaches to effectively and efficiently address the privacy issues and limit the leakage of identifiable information when the verification logic is outsourced to an untrustworthy environment. This talk also gives an overview of some of the most advanced Secure Signal Processing mechanisms recently proposed for this purpose, with an special attention on homomorphic encryption and lattice cryptography, their advantages and limitations, and the challenges posed by the marriage of cryptography and signal processing for privacy preservation in secure face verification scenarios.
Talk given at Biométrie, Indexation multimédia et Vie privée, Télécom ParisTech