Encryption scheme for low-altitude traffic images via learning-based fractal compression and bidirectional diffusion

With the development of intelligent transportation, low-altitude remote sensing images are increasingly used for traffic monitoring and detection. However, these high-volume images often contain sensitive information, and their leakage can seriously threaten privacy and security, motivating the need for joint compression and encryption. To address the joint compression and encryption requirements for low-altitude traffic images in privacy-preserving and resource-constrained scenarios, this paper proposes a solution that integrates deep learning-based fractal compression with efficient reversible diffusion. First, we construct a fractal compression network named FCnet, featuring an encoder–decoder architecture, which directly maps the original image to compact fractal codes. The FCnet fundamentally replaces the exhaustive block-matching search of traditional fractal encoding, while preserving the classical fixed-point decoding mechanism and its multi-scale reconstruction property. Subsequently, we design a reversible parallel bidirectional chained diffusion algorithm to encrypt compressed fractal codes. It leverages chaotic sequences to drive algebraic perturbations, dynamic generation of odd weights, and bit-level cyclic left-shift, achieving strong randomness and global diffusion effects. Extensive experiments demonstrate that the proposed scheme achieves superior performance in terms of security and efficiency, and supports on-demand multi-resolution decoding, making it well-suited for resource-constrained applications.