Deep Optical Cryptography: Joint Encryption and LC Lens Key Generation for Secure Displays

We present a learning‐based visual cryptography system that performs computational encryption and distance‐separated optical decryption using a liquid crystal (LC) modulator. The model jointly learns a ciphertext image and an LC‐based optical key such that their optical transmittance modulation recovers the plaintext at a designated capture distance. Decryption succeeds only for the correctly matched ciphertext–key pair, whereas mismatched keys produce visually uninformative outputs. We further address practical misalignment by incorporating shift‐aware training and a patch encryption strategy to enhance lateral ( X – Y ) shift tolerance. Quantitative evaluation using YOLO‐based digit recognition metrics and MS‐SSIM demonstrates improved robustness under spatial displacement, with experimental results consistent with simulation. These results validate a device‐bound, physically realizable visual cryptography framework that is promising for secure display, authentication, and anticounterfeiting applications.