A noise-constrained lightweight high-quality image hiding method based on invertible neural networks

Image steganography is an important branch of steganography that aims to hide secret information in a cover image, making it difficult to be detected by the human eye, while allowing only authorized parties to recover the secret information. Most existing image hiding methods have a large number of parameters, complex structures, and significant storage requirements. Although some methods have fewer parameters, they often produce low-quality results. To address these issues, this paper proposes a noise-constrained lightweight high-quality image hiding method based on invertible neural networks, called NCL-Net. The method introduces a novel multi-scale dense connection structure (MSDCS), an efficient interaction block (EIB), and a noise-constrained loss, achieving high capacity, high quality, lightweight design, and strong generalization in image hiding tasks. Extensive experiments demonstrate that the proposed method outperforms state-of-the-art methods in terms of generated image quality, model storage usage, lightweight design, and practicality.