Abstract
Face recognition technology is increasingly prevalent, yet securing facial image data remains a critical challenge due to privacy risks. This study introduces an innovative cryptosystem that utilizes optical interference and phase truncation theory to encrypt facial images, ensuring their secure transmission and storage. The system incorporates a dual-key mechanism, providing authorized users with the flexibility to decrypt specific or all images as needed. Decrypted images are subsequently used for face recognition, leveraging deep learning for enhanced accuracy. The proposed system exhibits strong resistance to various attacks while maintaining high computational efficiency. A key innovation is the Amplitude-Phase Separation Asynchronous Encryption (APSAE) technique, which mitigates inherent vulnerabilities by separately and asynchronously encrypting the amplitude and phase components. Experimental evaluations on the Labeled Faces in the Wild (LFW) dataset demonstrate a face recognition accuracy of [Formula: see text] on decrypted images, with encryption taking 40.47 s for 10,000 images and decryption requiring just 0.02 s per image. This research not only addresses the critical issue of facial image privacy leakage but also contributes to the advancement of secure biometric recognition systems, offering a promising avenue for future research and development in the field of data security.