Stable Hash Generation for Efficient Privacy-Preserving Face Identification

The development of large-scale facial identification systems that provide privacy protection the enrolled subjects represents an open challenge. In context protection, several template schemes have been proposed in past. However, these appear to be unsuitable for indexing (workload reduction) biometric systems. More precisely, they utilized performing exhaustive searches, thereby leading degradations computational efficiency. this work, we propose a privacy-preserving face system which utilisers Product Quantization-based hash look-up table and retrieval protected templates. These templates are through fully homomorphic encryption schemes, guaranteeing high subjects. For best configuration, experimental evaluation carried out over closed-set open-set settings shows feasibility technique use systems: workload reduction down 0.1% baseline approach search is achieved together with low pre-selection error rate less than 1%. terms performance, False Negative Identification Rate (FNIR) range 0.0% - 0.2% obtained practical Positive (FPIR) values on FEI FERET databases. addition, our proposal competitive performance unconstrained databases, e.g., LFW database. To authors’ knowledge, first work presenting scheme performs comparisons encrypted domain.