HL-IEM: a Hilbert-Logistic dual-layer chaotic encryption mechanism for lightweight and secure image communication

This study presents an advanced AI-Adaptive Hilbert and Logistic Image Encryption Mechanism (HL-IEM) that combines an Adaptive Feed-Forward Neural Network (FNN), a Logistic Chaotic Map, and Hilbert Curve Block Scrambling to achieve robust, lightweight, and content-aware image security. The FNN operates on image features such as mean intensity, variance, entropy, and edge density, and accordingly generates optimised chaotic parameters that control both pixel-level diffusion via the Logistic Map and adaptive block-level permutation via the Hilbert Curve. This collaboration between AI-driven chaotic control and nonlinear permutation-diffusion ensures strong key sensitivity, high randomness, and complete decorrelation of adjacent pixels. Experimental evaluation across standard test images (Peppers, Tree, Airplane, and House) demonstrates near-ideal information entropy (7.9488-7.9932), correlation coefficients close to zero after encryption, NPCR above 99.83%, and UACI around 32.90 to 33.78%. The MSE values range from 8125 to 8205, and the PSNR values range from 9.03 to 9.08 dB, indicating very high image quality. The encryption and decryption times are around 1 second for [Formula: see text] images, indicating that the technique can be used in real-time scenarios in IoT and multimedia environments. The proposed method increases computational load only slightly and has been confirmed to be quite resistant to statistical, differential, and brute-force attacks. These findings establish the AI-adaptive HL-IEM as a highly secure, efficient, and scalable image encryption solution suitable for both civilian and defence-grade communication systems.