Hybrid Cryptosystem for Image Encryption Using Lorenz Attractor and Neural Network Optimization

Abstract

Image encryption is required to protect visual data from unauthorized access. This study proposes a hybrid image cryptosystem based on the Lorenz chaotic attractor combined with artificial neural network optimization for adaptive chaotic parameter generation. The proposed method applies chaotic pixel permutation and bidirectional diffusion using chained XOR and bit rotation operations to improve ciphertext randomness and differential attack resistance. Experiments were conducted on three categories of generated RGB images with a resolution of 1024×1024 pixels. Performance evaluation was carried out using entropy, correlation coefficient, NPCR, UACI, PSNR, SSIM, key sensitivity, key space, histogram analysis, and encryption time analysis. The experimental results show that both encryption methods achieved entropy values close to the theoretical maximum approx ≈7.9999 and correlation values near zero, indicating strong randomness and successful removal of spatial pixel relationships. The proposed system also achieved NPCR values of approximately 99.6% and UACI values close to 333%, demonstrating strong resistance against differential attacks. In addition, the decryption process successfully reconstructed the original images without information loss, producing infinite PSNR values and SSIM values of 1.0. The obtained key sensitivity values exceeding 99.6% and the approximate key space of 〖10〗^45 further indicate strong dependence on secret key precision and resistance against brute-force attacks. Overall, the proposed hybrid cryptosystem demonstrated strong statistical security, stable computational performance, and effective encryption capability for high-resolution RGB images.