A robust and secure immensely random GAN based image encryption mechanism

Protection of information (data/images) is crucial in the colossal and ever-expanding domain of digital transfer. Cryptography is one of the well-known admired solutions to preserve images' confidentiality over highly unreliable and unrestricted public media. Researchers propose numerous techniques to accomplish the ever-growing need for security. In continuation, this paper aims to develop a robust image encryption scheme that accomplishes the task of protection by employing a series of specially designed substitution box, permutation box and diffusion box by taking encryption keys as input which is consequently generated from a Generative Adversarial Network (GAN), an unsupervised deep learning algorithm trained on the Logistic Maps. The substitution box performs byte-level substitution using two different schemes, and the other two perform encryption at both bit-level and byte-level, which helps it withstand a wide range of attacks. A dataset with 789 standard images is taken for experimentation, partitioned into three sets according to size (128, 256, and 512). The projected scheme outperforms state-of-the-art methods with better performance since the trained generator passed the comprehensive tests; it also withstands most of the probable attacks available in the literature. GAN was subjected to the chi-square test, runs test, and NIST test suite to check the randomness of the Pseudo-Random Number Generator. The projected algorithm offers promising visual, statistical, robustness, and quantitative analysis results.

Automated summary

This paper proposes an image encryption scheme based on Generative Adversarial Network, utilizing specially designed substitution box, permutation box, and diffusion box for confidentiality. The experiment shows that the proposed scheme outperforms state-of-the-art methods in terms of performance and resistance to attacks.