QryptGen+: a quantum GAN-based high-security image encryption key generator with enhanced chaotic modeling

Various approaches have been proposed to enable quantum generative adversarial networks (QGANs) to learn data distributions and generate images similar to original data. While QGANs are theoretically capable of modeling highdimensional, complex distributions with fewer parameters and shorter training times than classical GANs, current hardware limitations hinder their performance. As a result, research has increasingly focused on identifying specific domains where QGANs can offer unique advantages. In our prior work, we proposed QryptGen, a PQWGAN-based framework for generating 28 × 28 grayscale encryption key images from chaotic data. This demonstrated that QGANs can learn from visually indistinct, high-entropy distributions—beyond conventional datasets like MNIST—and produce image-based keys suitable for secure domains such as military communication, medical diagnostics, and cloud privacy systems. However, QryptGen’s use of row-wise patch stacking introduced inter-row correlations, reducing the randomness of the generated keys. Moreover, the PQWGAN loss function, optimized for structured data, was insufficient to fully capture the irregularity of chaotic distributions. To address these issues, we introduce QryptGen+, a redesigned framework that enhances randomness in key image generation. Key improvements include larger patch-wise generation to reduce structural bias, a strongly entangled quantum ansatz, a balanced training ratio between the generator and the critic, and a novel loss function that promotes anti-correlation and maximizes entropy. Through these refinements, QryptGen+ yields encryption keys with higher statistical security, reaffirming the potential of quantum machine learning for cryptographic applications.