Visually meaningful image encryption based on 2D compressive sensing and dynamic embedding

Conventional image encryption methods typically encrypt images into noise-like images, which can provide a degree of security for the image, but the non-visually meaningful appearance can easily trigger the suspicion of hackers. To address this issue, visually meaningful image encryption (VMIE) which can produce a visually meaningful cipher image is developed. Nevertheless, current VMIE algorithms still exhibit certain limitations in embedding capacity and the quality of the cipher images. To solve the above issues, a novel VMIE algorithm based on the improved sine-tangent (IST) map, two-dimensional compressive sensing (2DCS), and dynamic embedding is proposed in this paper. First, to enhance the embedding capacity of the proposed algorithm, 2DCS is applied to the plain image to obtain the compressed data, wherein the measurement matrix is generated by an IST map to reduce the use of storage space and optimized by the simulated annealing algorithm to reduce the distortion of the decrypted image. Then, scrambling and diffusion operations are applied to the compressed image to disguise its statistical properties. Finally, to adapt to plain images with different texture features and enhance the quality of VMIE, a dynamic embedding algorithm is devised to embed the encrypted image into the host image. Simulation results and robustness analysis confirm that our scheme is feasible. Additionally, the proposed algorithm shows the superior visual quality and reconstruction quality compared to existing related schemes.