Biological Template Protection Based on Multikey Cryptographic Decomposition Features

As more and more privacy protection mechanisms are introduced, facial recognition is gradually being optimized in terms of homo-morphic computing power. However, certain aspects have been overlooked, such as the reversibility of facial features due to key leakage and the degradation of accuracy when keys are exchanged to update the template library. To address this problem, we propose a feature grouping method that aims to improve the efficiency of ciphertext operation. This approach involves a multi-key encryption strategy for facial features that ensures their irreversibility. Experimental validation supports both the irreversibility and the ability to update within this framework. When rigorously tested, this scheme shows increased security compared to single-key frameworks because part of the key is withheld, making it difficult for attackers to obtain comprehensive facial feature data. In addition, even when using keys of the same length, attackers are faced with increased time expenditure. With regard to the upgradability of systems, experimental findings have shown for the first time that the loss of accuracy associated with homomorphic encryption algorithms is within tolerable limits. It is noteworthy that our proposed method has an even smaller loss in accuracy. When assessing the performance and precision of the algorithm, we achieved an impressive accuracy rate of 97.425% with processing times of 0.302 seconds per face. The benefits of our feature grouping method become more clear as the face template library grows. It outperforms non-grouping methods by more than 4 seconds on nearly 1,000 face libraries.