URSMamba: Universal remote sensing image steganography using state space model.

Image steganography involves concealing a secret image within a cover image for covert transmission, which has been widely developed in information security. However, current research often overlooks remote sensing images, which inherently possess higher confidentiality. The abundant ground object distributions and multiple spectral bands in remote sensing images present significant challenges to concealing and revealing ability, and hiding capacity of image steganography. In this work, we propose URSMamba, a universal remote sensing image steganography based on State Space Model. Specifically, we introduce a Low-High Frequency Mamba Block (LHfreMB) that jointly captures global representations and local details to model the complex ground distributions and object features in remote sensing images effectively. Then, rich spectral information is extracted by Spectral Mamba Block (SpectralMB). The Spatial-Spectral Dynamic Fusion (SSDF) block is proposed to adaptively guide the dynamic fusion of spatial and spectral features based on their importance, effectively integrating spatial and spectral information during concealing and revealing. We achieve superior performance on remote sensing images in multiple spectral bands, while delivering high-quality results on natural images. Compared with existing state-of-the-art methods, the proposed URSMamba improves the PSNR of cover/stego and secret/recovery image pairs by 0.54 dB and 1.65 dB when hiding 8-band, and by 3.25 dB and 2.96 dB on natural images. The code is available at https://github.com/Floren1026/URSMamba.