Camouflaged image cryptosystem based on compressive interference against deep learning

In this paper, our previously proposed compressive interference-based image cryptosystem (CIBC) is utilized to develop a camouflaged image cryptosystem, in which a secret image is camouflaged by a meaningful image. During the encryption process, a secret image is encoded into two sparse and complementary phase only masks (POMs) by CIBC under the constraints of two binary amplitude masks(BAMs), in which the BAM can be taken as a private key. Then, a camouflaged image is encoded into another two POMs by CIBC under the constraints of the two POMs generated in the aforementioned step, and the two final POMs are stored and transmitted as ciphertexts. In the decryption process, only authorized users, who can access the ciphertexts, public and private keys simultaneously, can retrieve the decrypted secret image. Otherwise, even if the eavesdropper cracks the ciphertexts and the public keys, only the camouflaged image can be directly obtained, which can confuse the eavesdropper and then protect the secret image. Several simulation experiments were demonstrated to verify the feasibility and security of this cryptosystem.