A Joint Image Encryption and Compression Algorithm Based on CycleGAN

Digital images are easy to copy, spread, and tampering, therefore, protecting image contents in the digital era has emerged as a key focus of multimedia security research. Recently, researchers have introduced deep learning technology into cryptography, but how to reasonably integrate cipher keys into neural networks and how to combining compression with encryption are still a challenge work. This paper proposes a new image encryption-compression network based on CycleGAN, which is strictly in accordance with the principle of cryptography. In the encryption phase, the key and the plain image are fed into the encryption network to generate a one-channel cipher image. Conversely, in the decryption phase, the same key and the cipher image are fed into the decryption network to perform deciphering, while the cipher image cannot be deciphered when the incorrect key is used. Both encryption and decryption networks take use of CycleGAN, and a new auxiliary network is designed to ensure the key sensitivity to the encryption-compression network, thus the security of the system is enhanced. Experimental results have demonstrated the feasibility of the proposed scheme.