Scholarlink

<p>This paper proposes an innovative deep learning framework incorporating chaotic permutation for zero-watermark tasks. Conventional zero-watermark approaches that rely on handcrafted features demonstrate critical limitations under complex attack scenarios, such as noise, compression, and rotation. It may stem from their inability to extract discriminative features that effectively preserve robustness against diverse attack modalities. To overcome these limitations, a deep learning framework is implemented to automatically extract semantic features, thereby significantly enhancing the intrinsic robustness of digital copyright protection. Despite the improvement in robustness, the feature extraction process remains vulnerable to security breaches, especially in scenarios involving interception attacks. To address this issue, we introduce a chaotic permutation mechanism. Through the inherent unpredictability of chaotic systems, this process significantly enhances the complexity and security of the generated zero-watermark. Extensive experimental results demonstrate the effectiveness of our proposed method. We achieve an average normalized correlation of 0.98 across different types of attacks, based on comparisons between watermarks extracted from attacked images and the original watermark. These results highlight the particular resilience against different types of attacks. The source code and details are available at <a href="https://github.com/JackPeng007/ZeroWatermark.">https://github.com/JackPeng007/ZeroWatermark.</a></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A robust zero-watermark method based on deep learning and chaotic permutation

  • 期刊论文

  • 发表日期：

    2025年6月8日

查看全文
  • Yiran Peng,
  • Qingqing Hu,
  • Jing Xu,
  • Cuiyun Lin,
  • Chenheng Deng,
  • Yiyao Huang,
  • Wenxiao Wang,
  • Kintak U

摘要

This paper proposes an innovative deep learning framework incorporating chaotic permutation for zero-watermark tasks. Conventional zero-watermark approaches that rely on handcrafted features demonstrate critical limitations under complex attack scenarios, such as noise, compression, and rotation. It may stem from their inability to extract discriminative features that effectively preserve robustness against diverse attack modalities. To overcome these limitations, a deep learning framework is implemented to automatically extract semantic features, thereby significantly enhancing the intrinsic robustness of digital copyright protection. Despite the improvement in robustness, the feature extraction process remains vulnerable to security breaches, especially in scenarios involving interception attacks. To address this issue, we introduce a chaotic permutation mechanism. Through the inherent unpredictability of chaotic systems, this process significantly enhances the complexity and security of the generated zero-watermark. Extensive experimental results demonstrate the effectiveness of our proposed method. We achieve an average normalized correlation of 0.98 across different types of attacks, based on comparisons between watermarks extracted from attacked images and the original watermark. These results highlight the particular resilience against different types of attacks. The source code and details are available at https://github.com/JackPeng007/ZeroWatermark.