A sequential image encryption based on random neural permutation and reversible cellular automata
Year: 2026
Type: article
Abstract: This study presents a keyed Random Neural Permutation Layer–Reversible Cellular Automata (RNPL–RCA) image-encryption framework that integrates neural permutation, reversible cellular automata, whitening operations, cross-channel diffusion, and final keyed diffusion. In the proposed method, the RNPL module generates key-dependent permutation masks for red–green–blue (RGB) channel scrambling, while the reversible sum–XOR RCA module provides invertible diffusion. Additional intra-channel and cross-channel diffusion stages strengthen confusion, diffusion, and avalanche behavior. Experiments on benchmark color, grayscale-style, medical, texture, and lunar surface images demonstrate that the proposed method produces cipher images with near-uniform histograms, an entropy of 7.9989 bits, a near-zero adjacent-pixel correlation coefficient, a Number of Pixels Change Rate (NPCR) of 99.6302%, and a Unified Average Changing Intensity (UACI) of 33.4941%. The results further demonstrate strong key sensitivity and resistance to known-plaintext and chosen-plaintext attacks, together with satisfactory performance in randomness-based security tests. The decrypted images are reconstructed losslessly, achieving a Mean Absolute Error (MAE) of 0, a Peak Signal-to-Noise Ratio (PSNR) of ∞, and a Structural Similarity Index (SSIM) of 1. These results demonstrate that the proposed keyed RNPL–RCA framework provides a robust, reversible, and fidelity-preserving approach to image encryption.
Source: Research in Mathematics
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7225424957