Adaptive Image Encryption Using Artificial Intelligence and Crypto-GANs
Abstract
Crypto-Generative Adversarial Networks (Crypto-GANs) as an intelligent and adaptive framework for secure image encryption. By combining adversarial deep learning with cryptographic principles, the proposed system learns complex non-linear transformations to convert plain images into noise-like cipher images while enabling accurate reconstruction using the correct secret key. Unlike traditional encryption methods based on fixed mathematical rules, Crypto-GANs dynamically adapt to emerging attack strategies. Experimental results demonstrate strong security performance. Encrypted images achieved entropy values close to the theoretical maximum (up to 7.9978), indicating high randomness. Adjacent pixel correlations were reduced to near zero, confirming the removal of inherent statistical patterns. High NPCR and UACI values reflected effective diffusion, ensuring resistance against differential and chosen-plaintext attacks. Decryption quality remained excellent, with high PSNR and SSIM values supporting accurate image recovery.