QryptGen: a quantum GAN-based image encryption key generator using chaotic data distributions

Abstract

The emergence of generative adversarial networks (GANs) has led to tremendous advancements in deep learning-based AI for image generation. While many researchers have used GANs to generate human faces and numeric images, others have applied them to learn and generate images from data distributions created by less distinct, chaotic systems. These chaotically generated images can serve as encryption keys for simple image encryption methods, potentially useful in military applications where data security is crucial, or in hospitals handling sensitive images like X-rays, CTs, MRIs, and physical photographs. Meanwhile, quantum GANs are still in their early research stages, primarily learning from distinct images like those in the MNIST or Fashion MNIST datasets. In this paper, we demonstrate that quantum machine learning models, specifically QGANs, can also learn from non-descript chaotic data distributions. We propose QryptGen (quantum crypt generator), which produces 28 \(\times \) 28 pixel grayscale image encryption keys. We show that encryption keys generated through quantum machine learning techniques can achieve a level of security comparable to those generated by classical deep learning techniques, thus confirming the potential of quantum machine learning to contribute broadly beyond just image encryption. Specifically, our study employs patch QGAN with a minimal number of qubits to maximize quantum advantages on NISQ devices, enhancing practicality.