Deep face fuzzy vault: Implementation and performance

Biometric technologies, especially face recognition, have become an essential part of identity management systems worldwide. In deployments biometrics, secure storage biometric information is necessary in order to protect the users’ privacy. this context, cryptosystems are designed meet key requirements protection enabling a privacy-preserving and comparison data, e.g. feature vectors extracted from facial images. Until now, hardly been applied state-of-the-art recognition utilizing deep convolutional neural networks. This work investigates application well-known cryptosystem, i.e. improved fuzzy vault scheme, through To end, transformation method introduced which maps fixed-length real-valued integer-valued sets. As said transformation, detailed analysis different quantisation binarisation techniques conducted. At binding, obtained sets locked unlinkable vault. For retrieval, efficiency polynomial reconstruction investigated. The proposed template scheme agnostic characteristic and, thus, can be virtually any features computed by network. experiments, vault-based constructed employing with network trained additive angular margin loss (ArcFace). best configuration, false non-match rate below 1% at match 0.01%, achieved cross-database experiments on FERET FRGCv2 databases. On average, security level up approximately 28 bits obtained. presents effective face-based providing privacy reference data as well digital derivation face.