About the journal
Browse
Collections
Multimedia collections
Authors & reviewers
Multidimensional Message Processing with Texture-Guided Priors for Robust Bit-Level Image Watermarking
Ruohan Meng , Jiaqi Chen , Xingyuan Zhao , Baowei Wang , Zhangjie Fu
Robust bit-level image watermarking requires a careful balance between message recoverability under distortions and the perceptual fidelity of watermarked images. Existing message processors typically transform a one-dimensional payload into spatial feature maps without explicitly considering the structural characteristics of the carrier image, which may lead to inefficient redundancy allocation and structured visual artifacts. To address this limitation, we propose MDTG, a carrier-aware watermarking framework equipped with a two-stage message processor. The Multidimensional Message Processor constructs complementary message representations through reshaping, global mixing, and duplication branches, whereas the Texture-Guided Message Processor progressively injects multi-scale high-frequency carrier priors into the message features. We further introduce a multi-scale perceptual objective that jointly constrains feature reconstruction, texture consistency, and wavelet- domain fidelity. Experiments on COCO with 64-bit payloads demonstrate that MDTG achieves 57.65 dB PSNR in the distortion-free setting and 0.318% BER under combined distortions, while maintaining robustness to diverse common and unseen image transformations. Additional analysis reveals a fidelity-robustness trade-off under aggressive JPEG compression.
Deep image watermarking / bit-level watermarking / message processing
Higher Education Press 2026
/
| 〈 |
|
〉 |