Optical color ghost cryptography and steganography based on multi-discriminator generative adversarial network

An optical color ghost cryptography and steganography based on multi-discriminator generative adversarial network (MDGAN) is proposed. The system comprises a digital light projector configured to project color patterns onto a secret image and a bucket detector for collecting the light intensity in the output plane, rather than using three basic independent color channels. By applying a non-obvious data integration process, the detected bucket signals are hidden into those of a non-secret image. It can be shown that the decrypted images with conventional decryption methods always have serious noise and distortion. In this method, the periphery and detail of the secret images are reconstructed by using a pretrained MDGAN, where a multi-discriminator is employed to achieve a better image understanding. By continuously optimizing the parameters of the generator network, a high quality of image reconstruction can be achieved even if the sampling rate is extremely low. The feasibility of the proposed method is demonstrated by numerical simulations.