Image Encryption Based on 2D Discrete Hyperchaotic Hopfield Neural Network and Pixel Sbox
Caizheng Liu, Ning Lin, Tao Wang, Hongwei Zhang, Zongli Yang, Dawei Ding · 2025
Among the current image encryption methods, the traditional S-box can only achieve byte-level pixel replacement, resulting in low efficiency. Moreover, independent encryption of RGB channels leaves behind statistical features, such as strong correlation between channels. This paper proposes a pixel-level S-box encryption framework based on hyperchaotic discrete Hopfield neural networks. Design a twodimensional discrete Hopfield neural network with hyperchaotic characteristics to generate highly random sequences. Innovatively construct the pixel S-box, directly replacing the entire pixel with the pixel values of the $\mathbf{R}$ and $\mathbf{G}$ channels as the coordinate index to overcome the efficiency limitation of byte-level replacement. Flatten the three-channel image into a grayscale image, and use chaotic sequences for cross-channel pixel scrambling and confusion to eliminate the statistical correlation between channels. The key space of this scheme is quite large and can fend off brute-force cracking. Via experiments with encrypted images, it turns out that the encryption works well.