Spatiotemporal Chaos in Exponential Cosine Polynomial Nonlinear Coupling and its Application in Adaptive Multi-Image Encryption with Autoencoder

Spatiotemporal chaotic systems possess high dimensional dynamics and strong spatial and temporal diffusion capabilities, making them highly suitable for secure communication. However, many existing models suffer from limited chaotic regions and insufficient security. To address these issues, a novel spatiotemporal chaotic system featuring exponential cosine polynomial nonlinear coupling and dynamic nonlocal interactions is constructed as the core of an adaptive multi-image encryption scheme. The proposed model, named Exponential Cosine Polynomial Coupled Map Lattice (ECPCML), enhances chaotic behavior through nonlocal coupling and cosine-based coefficients with dynamic adjustment, resulting in a wider chaotic region and higher unpredictability. To further enhance security, a 3D Knight’s Tour Scrambling (3D-KTS) algorithm is employed to realize cross channel scrambling along 24 movement directions, followed by Affine Pixel Diffusion (APD) for pixel level protection. An autoencoder is integrated for lossy compression to improve transmission efficiency without significantly degrading image quality. Additionally, a multi-image fusion strategy enables simultaneous encryption of multiple images with varying sizes and channels. Experimental results and comparative analyses confirm that the proposed scheme achieves favorable performance in security, compression efficiency, and resistance to chosen-plaintext attacks, demonstrating its potential for secure image transmission in complex network environments.