A Novel Image Encryption Based on Style Transfer
Wanyi Zhou, Yujin Lu, Rui Wang, Qi Wang · 2023
This paper introduces an image encryption scheme that combines a style transfer model with chaotic encryption, offering the adaptability of deep learning encryption and the randomness of chaos. During style transfer, a noise image is creatively treated as the style image, while the plaintext image serves as the content image, resulting in a Noise-Style image. Style features of the plaintext are extracted for decryption. The Logistic map is found to be effective in encrypting Noise-Style images, and the Henon map enhances security. The algorithm's key consists of style features and a chaotic key. This approach integrates deep learning into traditional chaotic encryption, making it more complex and secure. Numerical simulations and experiments confirm its feasibility and effectiveness, especially in resisting noise attacks, highlighting its robustness.