Robust and discriminative zero-watermark scheme based on invariant feature and similarity-based retrieval for protecting large-scale DIBR 3D videos

Xiyao Liu, Yifang Wang, Ziqiang Sun, Lei Wang, Rongchang Zhao, Yuesheng Zhu, Beiji Zou

Abstract

Digital rights management (DRM) of depth-image-based rendering (DIBR) 3D video is an emerging area of research. Existing schemes for DIBR 3D video cause video distortions, are vulnerable to severe signal and geometric attacks, cannot protect 2D frame and depth map independently or can hardly deal with large-scale videos. To address these issues, a novel zero-watermark scheme based on invariant feature and similarity-based retrieval for protecting DIBR 3D video (RZW-SR3D) is proposed in this study. In RZW-SR3D, invariant features are extracted to generate master and ownership shares for providing distortion-free, robust and discriminative copyright identification under various attacks. Different from traditional zero-watermark schemes, features and ownership shares are stored correlatively, and a similarity-based retrieval phase is designed to provide effective solutions for large-scale videos. In addition, flexible mechanisms based on attention-based fusion are designed to protect 2D frame and depth map independently and simultaneously. Experimental results demonstrate that RZW-SR3D have superior DRM performances than existing schemes.

Introduction: The Challenge of 3D Video Protection

Protecting 3D video content is significantly more complex than protecting standard 2D video. Modern 3D content typically utilizes Depth-Image-Based Rendering (DIBR), where a scene is represented by a 2D color frame and a corresponding depth map. This format allows for efficient transmission but introduces unique vulnerabilities for Digital Rights Management (DRM).

Existing DRM solutions face three primary hurdles:

  1. Visual Distortion: Traditional watermarking techniques embed data directly into pixel values, which can perceptibly degrade the 3D viewing experience.
  2. DIBR Invariance: Synthesizing new 3D views involves horizontal pixel shifts. Most watermarks are destroyed during this geometric transformation.
  3. Scalability Issues: "Zero-watermark" schemes, which map video features to an external watermark, often fail in large-scale libraries because they lack efficient mechanisms to match a queried video to its specific ownership record in a database.

The RZW-SR3D Framework

To overcome these limitations, the researchers developed RZW-SR3D (Robust Zero-Watermark with Similarity-based Retrieval for 3D). This scheme identifies ownership without adding a single bit of distortion to the video content.

Comparison of traditional watermarking (a, b), traditional zero-watermarking (c), and the proposed RZW-SR3D scheme (d).

As shown in Figure 1, the proposed scheme (d) differs from traditional approaches by integrating a similarity-based retrieval phase. This allows the system to precisely locate the relevant ownership shares even when the specific identity of the queried video is unknown or "doubtful" in a large database.

Core Methodology

The workflow is divided into three distinct phases:

The three-phase workflow of the RZW-SR3D scheme: registration, retrieval, and identification.
  1. Feature Extraction: The system calculates the centroids of normalized deviations between Temporally Informative Representative Images (TIRI) and individual frames. A concentric ring partition method is applied, making these features invariant to rotation and flipping—common attacks used to bypass DRM.
  2. Zero-Watermarking via VSS: Using (2,2) Visual Secret Sharing (VSS), the system generates a "master share" from video features and an "ownership share" linking those features to the watermark. Only the ownership share is stored externally, ensuring the video remains lossless.
  3. Similarity-Based Retrieval: This is the critical component for scalability. When a suspicious video is queried, the system performs a similarity search in the feature database to retrieve the correct ownership share for verification.
  4. Attention-Based Fusion: Because 2D frames and depth maps have different textures and characteristics, an attention-based mechanism fuses their features. This enhances reliability by prioritizing the more robust component under specific types of attacks.

Experimental Evaluation

The study utilized a database of 200 DIBR 3D video clips from sources including the MPEG 3DAV group and Microsoft Research. The system was tested against 26 different attack types, including Gaussian blurring, rotation, cropping, and frame dropping.

Key Results

  • Retrieval Accuracy: The false-negative rate ($P_{fn}$) was remarkably low, averaging 0.0029 when fusing 2D and depth components.
  • Watermark Recovery: The Bit Error Rate (BER) for recovered watermarks was significantly lower than state-of-the-art schemes. Under fusion, the average BER was just 0.0287, compared to 0.1759 for competing AUEW methods.
  • Robustness: Recovered watermarks remained clearly recognizable even after heavy noise or 90-degree rotations, scenarios where traditional methods typically fail.

Conclusion and Future Work

RZW-SR3D provides a robust, lossless, and scalable solution for 3D video DRM. By moving the watermark out of the video stream and into a searchable feature database, it preserves visual quality while ensuring strong copyright protection. Future research will explore extending this invariant feature approach to protect medical volume data (e.g., 3D CT or MRI scans), where data integrity is paramount.

Summary updated

Abstract

Digital rights management (DRM) of depth-image-based rendering (DIBR) 3D video is an emerging area of research. Existing schemes for DIBR 3D video cause video distortions, are vulnerable to severe signal and geometric attacks, cannot protect 2D frame and depth map independently or can hardly deal with large-scale videos. To address these issues, a novel zero-watermark scheme based on invariant feature and similarity-based retrieval for protecting DIBR 3D video (RZW-SR3D) is proposed in this study. In RZW-SR3D, invariant features are extracted to generate master and ownership shares for providing distortion-free, robust and discriminative copyright identification under various attacks. Different from traditional zero-watermark schemes, features and ownership shares are stored correlatively, and a similarity-based retrieval phase is designed to provide effective solutions for large-scale videos. In addition, flexible mechanisms based on attention-based fusion are designed to protect 2D frame and depth map independently and simultaneously. Experimental results demonstrate that RZW-SR3D have superior DRM performances than existing schemes. First, RZW-SR3D can extracted the ownership shares relevant to a particular 3D video precisely and reliably for effective copyright identification of large-scale videos. Second, RZW-SR3D ensures lossless, precise, reliable and flexible copyright identification for 2D frame and depth map of 3D videos.
BibTeX
@misc{liu2017robust,
  author = {Xiyao Liu and Yifang Wang and Ziqiang Sun and Lei Wang and Rongchang Zhao and Yuesheng Zhu and Beiji Zou},
  title = {{Robust and discriminative zero-watermark scheme based on invariant feature and similarity-based retrieval for protecting large-scale DIBR 3D videos}},
  year = {2017},
  eprint = {1712.09480},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1712.09480},
}
Figure