Machine Learning-Optimized Hybrid Chaotic Map for Secure Image and Multimedia Encryption in IoT

Abstract This study proposes a Machine Learning-Optimized 3D Hybrid Chaotic Map (Logistic–Tent–Hénon–Chebyshev) for secure and energy-efficient encryption of images and multimedia data in resource-constrained IoT and edge computing environments. Four classical chaotic maps—Logistic, Tent, Hénon, and Chebyshev—are coupled using modular arithmetic to construct a novel 3D hybrid chaotic system. The parameters of this hybrid map are dynamically optimized by integrating Non-dominated Sorting Genetic Algorithm-II (NSGA-II) with a Back-Propagation Neural Network. The network takes the SHA-256 hash of the plain image as input to generate adaptive initial conditions and control parameters. The encryption scheme employs SHA-256 key generation, pixel confusion via a generalized Arnold cat map, and dual diffusion using two independent chaotic sequences. The proposed method supports both static images and video frame sequences and is validated on embedded hardware platforms including Raspberry Pi 5 and Jetson Nano. Experimental results on standard benchmark images and video sequences demonstrate superior performance, achieving an information entropy of 7.9978, NPCR of 99.68%, UACI of 33.52%, near-zero correlation coefficients (≈0.001), and successful passage of all NIST statistical tests. For a 256×256 image, the scheme requires only 78 ms encryption time and real-time power consumption of 92 mW on Raspberry Pi 5 and Jetson Nano, significantly outperforming AES-256 and recent hybrid chaotic schemes in security, speed, and energy efficiency. Unlike existing approaches, the proposed method performs dynamic ML-based parameter optimization with real-time power evaluation on embedded IoT platforms.