Deep-learning-assisted optical steganographic encryption via ghost encoding and binary hologram

We propose a steganographic approach for encrypting optical images by integrating ghost encoding and computer-generated holography. The process involves encoding a secret image using ghost imaging during its collection and subsequently encrypting it into a binary amplitude-only hologram (BAOH) with steganographic characteristics. The resultant BAOH servers as the ciphertext and can be printed out on a polyethylene terephthalate film using an inkjet printer. To enhance flexibility and reconstruction quality, we incorporate a deep neural network driven by sparse and noisy data to work alongside into the cryptosystem. Both the pure digital hologram cyphertext and its physical representation can be successfully decrypted. Experimental results confirm the feasibility of our scheme, offering benefits such as easy-to-carry storage, low production costs, and high security.