Local Geometric Distortions Resilient Watermarking Scheme Based on Symmetry

Zehua Ma, Weiming Zhang, Han Fang, Xiaoyi Dong, Linfeng Geng, Nenghai Yu

Abstract

As an efficient watermark attack method, geometric distortions destroy the synchronization between watermark encoder and decoder. And the local geometric distortion is a famous challenge in the watermark field. Although a lot of geometric distortions resilient watermarking schemes have been proposed, few of them perform well against local geometric distortion like random bending attack (RBA). To address this problem, this paper proposes a novel watermark synchronization process and the corresponding watermarking scheme. In our scheme, the watermark bits are represented by random patterns. The message is encoded to get a watermark unit, and the watermark unit is flipped to generate a symmetrical watermark. Then the symmetrical watermark is embedded into the spatial domain of the host image in an additive way. In watermark extraction, we first get the theoretically mean-square error minimized estimation of the watermark. Then the auto-convolution function is applied to this estimation to detect the symmetry and get a watermark units map. According to this map, the watermark can be accurately synchronized, and then the extraction can be done. Experimental results demonstrate the excellent robustness of the proposed watermarking scheme to local geometric distortions, global geometric distortions, common image processing operations, and some kinds of combined attacks.

The Synchronization Challenge in Digital Watermarking

Digital watermarking—the process of embedding hidden data into images for copyright protection or authentication—faces a significant hurdle: synchronization. When an image undergoes geometric changes like rotation, scaling, or cropping (global distortions), the decoder often loses track of the watermark's spatial coordinates. While global distortions have been largely addressed by existing research, local geometric distortions remain a formidable challenge.

A prime example of such a distortion is the "Random Bending Attack" (RBA), a component of the Stirmark benchmark. Unlike global affine transformations that affect the entire image uniformly, RBA warps local regions unevenly. This localized warping breaks the synchronization of standard decoders that rely on global patterns or periodic tiling.

Visualization of the host image and the effects of a Random Bending Attack (RBA) on both the image and a reference grid.

Methodology: Shifting from Periodicity to Symmetry

Traditional watermarking schemes often use periodic tiling—repeating the same block across the image—and search for peaks in the Auto-Correlation Function (ACF) to resynchronize. However, local distortions warp these blocks inconsistently, making global recovery difficult. This paper proposes a fundamental shift: using symmetry instead of periodicity for synchronization.

Symmetrical Embedding

Instead of simply tiling a watermark unit $w_m$, the proposed scheme flips the unit horizontally and vertically to create a symmetrical pattern. This symmetrical watermark is then embedded into the spatial domain of the host image. To ensure the watermark remains imperceptible, an adaptive strength strategy is used, adjusting the embedding intensity based on local image variance.

Auto-Convolution and Corner Detection

To extract the watermark, the decoder first generates a theoretically mean-square error minimized estimation of the hidden signal. Rather than using ACF, which detects translation, the researchers employ the Auto-Convolution Function (ACNF) to detect centers of symmetry.

The peaks in the resulting ACNF map correspond directly to the corners of the watermark units. By identifying these specific corner points, the decoder can treat the distortion on each individual unit as a perspective transformation. This is a critical improvement over simple affine models, as perspective transformations have more degrees of freedom and can more accurately approximate the complex local warping found in attacks like RBA.

Statistical State Determination

Flipping a block creates eight possible orientations (states). To avoid wasting data capacity on a "reference watermark" to indicate orientation, the authors use a statistical approach. They apply a null hypothesis test and Kullback-Leibler Divergence (KLD) to determine which orientation most likely contains the spread-spectrum signal, allowing for "free" synchronization without extra overhead.

Experimental Evaluation

The scheme was rigorously tested using the USC-SIPI Image Database, the BOSSBase Database, and the Stirmark 4.0 benchmark.

Robustness to Attacks

  • Random Bending Attack (RBA): The proposed scheme demonstrated superior performance. At low-strength RBA, the Bit Error Quantity (BEQ) remained near zero. At high strengths (e.g., strength 1.0), the proposed method achieved a BEQ of approximately 0.6, significantly outperforming competing methods like V_ACF, which saw errors rise to ~9.3.
  • Cropping and Combined Attacks: The method remained robust even when 75% of the image was removed. It also maintained low error rates under combined attacks, such as RBA followed by JPEG compression or rotation.
  • Imperceptibility: The average Peak Signal-to-Noise Ratio (PSNR) was maintained at 41.35 dB, ensuring that the embedded watermark is invisible to the human eye.

Conclusion and Future Directions

By leveraging symmetry and the Auto-Convolution Function, this research provides a robust solution to the long-standing problem of local geometric distortion in watermarking. The use of perspective transformations for local synchronization allows for much higher recovery accuracy than traditional affine-based methods.

Future work may focus on improving computational efficiency, as symmetry calculations can be intensive for very high-resolution images. Additionally, the authors suggest this symmetry-based approach could be highly effective for print-camera scenarios, where physical lens distortion and perspective tilt create significant local geometric noise.

Summary updated

Abstract

As an efficient watermark attack method, geometric distortions destroy the synchronization between watermark encoder and decoder. And the local geometric distortion is a famous challenge in the watermark field. Although a lot of geometric distortions resilient watermarking schemes have been proposed, few of them perform well against local geometric distortion like random bending attack (RBA). To address this problem, this paper proposes a novel watermark synchronization process and the corresponding watermarking scheme. In our scheme, the watermark bits are represented by random patterns. The message is encoded to get a watermark unit, and the watermark unit is flipped to generate a symmetrical watermark. Then the symmetrical watermark is embedded into the spatial domain of the host image in an additive way. In watermark extraction, we first get the theoretically mean-square error minimized estimation of the watermark. Then the auto-convolution function is applied to this estimation to detect the symmetry and get a watermark units map. According to this map, the watermark can be accurately synchronized, and then the extraction can be done. Experimental results demonstrate the excellent robustness of the proposed watermarking scheme to local geometric distortions, global geometric distortions, common image processing operations, and some kinds of combined attacks.
BibTeX
@article{ma2021local,
  author = {Zehua Ma and Weiming Zhang and Han Fang and Xiaoyi Dong and Linfeng Geng and Nenghai Yu},
  title = {{Local Geometric Distortions Resilient Watermarking Scheme Based on Symmetry}},
  journal = {IEEE Trans. Circuits Syst. Video Technol.},
  year = {2021},
  url = {https://openreview.net/forum?id=P7JO0ubhoTd},
}
Figure