Intelligent Hybrid Image Encryption Using Adaptive Neural Permutation and Chaotic Key Dynamics

In the era of digital imagery, in critical applications like healthcare, defense and private communication, there is a true need for strong, computationally secure encryption techniques. Traditional cryptographic algorithms, though optimal for text data, are prone to difficulty with images due to high pixel interdependence and data redundancy.. This paper proposes a novel multi-layered image encryption framework that uses Neural Network integrated with multiple chaotic encryptions. It uses a content-adaptive permutation process, facilitated by a deterministic neural network that combined with a diffusion process using chaotic maps. This proposed methodology uses the plaintext sensitivity of the image combined with a user-specified cryptographic key to produce a unique pixel-shuffling permutation, effectively reducing pixel correlation. A chaotic transformation seeded with cryptographically secure keys then manipulates pixel values to produce far-reaching diffusion. This hybrid technique offers good resistance to statistical, differential and brute-force cryptanalysis and is hence a very effective solution for modern-day image security applications. Average test results from 10,224 encrypted images show high entropy = 7.9855, pixel correlation near zero i.e.-0.002 to 0.06, high MSE = 105.505, PSNR = 27.899 dB, low SSIM = 0.00820 and effective immune against differential attacks with NPCR = 99.60% and UACI = 50.0123%. The MAE = 127.453 guarantees strong obscuration making the proposed method highly appropriate for high-security applications.