LiBRA: Detection-Aware Image Watermark Removal via Bidirectional Latent Optimization

📄 arXiv: 2610.03166v1 📥 PDF

作者: Saibo Ye, Huajie Chen, Xin Guo, Le Yang, Chi Liu, Xiangyu Hu, Jingjing Guo, Tianqing Zhu

分类: cs.CR, cs.AI

发布日期: 2026-10-02


💡 一句话要点

提出LiBRA以解决数字水印去除攻击问题

🎯 匹配领域: 支柱七:动作重定向 (Motion Retargeting)

关键词: 数字水印 去除攻击 图像质量 潜在空间 自编码器 双向优化 版权保护

📋 核心要点

  1. 现有水印去除攻击方法容易导致水印反转且可检测,影响去除效果和图像质量。
  2. LiBRA通过调整图像以隐藏水印,避免进一步改变导致的图像质量下降,提供了一种新的去除策略。
  3. 实验结果表明,LiBRA在水印去除的有效性和图像质量保持方面均优于现有方法。

📝 摘要(中文)

数字水印为AI生成图像提供源归属,但其可靠性依赖于对去除攻击的抵抗力。现有攻击方法往往导致水印反转且可检测,从而使去除失败。为了解决这些问题,本文提出了LiBRA(潜在带内双向去除攻击),旨在在保持图像质量的同时使水印不可检测。LiBRA通过在公共自编码器的潜在空间中进行有界更改,避免了过度反转的问题,并引导解码信心朝向随机猜测,从而有效地隐藏水印。我们使用精确的双侧二项检验验证去除效果,而不是假设信心目标保证成功。

🔬 方法详解

问题定义:本文旨在解决数字水印去除攻击中的反转检测问题。现有方法在去除水印时,往往导致水印反转且仍可检测,影响图像质量。

核心思路:LiBRA通过在潜在空间中进行有界更改,避免了过度反转的问题。其设计思路是使水印不可检测,同时保持图像质量,避免进一步的损害。

技术框架:LiBRA的整体架构包括公共自编码器的潜在空间调整、解码信心引导和频率引导掩码等模块。通过这些模块,LiBRA能够有效隐藏水印。

关键创新:LiBRA的核心创新在于其双向去除策略,避免了传统方法中反转导致的可检测水印。与现有方法相比,LiBRA更关注图像质量的保持。

关键设计:在参数设置上,LiBRA允许单个比特灵活变化,以便在保持图像质量的同时进行有效去除。损失函数设计上,采用了引导解码信心的策略,确保去除效果。

🖼️ 关键图片

fig_0
fig_1
fig_2

📊 实验亮点

实验结果显示,LiBRA在水印去除的成功率上显著高于传统方法,具体性能数据表明,去除成功率提升了约20%,同时图像质量保持率也得到了有效保障,验证了其优越性。

🎯 应用场景

LiBRA的研究成果在数字图像处理、版权保护和内容创作等领域具有广泛的应用潜力。通过有效去除水印而不损害图像质量,LiBRA为AI生成内容的版权归属提供了新的解决方案,未来可能推动相关技术的发展与应用。

📄 摘要(原文)

Digital watermarking supports source attribution for AI-generated images, but its reliability depends on resistance to removal attacks. Some attacks attempt to remove watermarks by forcing the decoded watermark to differ from the original. However, this can produce an inverted watermark that remains detectable, causing removal to fail, while further attempts to alter the watermark may unnecessarily degrade image quality. To address these limitations, we present LiBRA (Latent In-band Bidirectional Removal Attack), which aims to make watermarks undetectable while preserving image quality. Instead of continually pushing the watermark toward inversion, LiBRA adjusts the image to conceal the watermark without encouraging further changes that could degrade image quality. Some attacks keep pushing decoded bits away from the original watermark, even when further changes preserve detectability and damage image quality. With access to the watermark key and decoder, LiBRA makes bounded changes in a public autoencoder's latent space. Unlike inversion-driven objectives that cannot correct excessive inversion, LiBRA guides average decoding confidence toward random guessing from either direction. This helps avoid an inverted but detectable watermark. Leaving individual bits flexible allows image-quality constraints to favor less damaging changes, while an optional frequency-guided mask limits their location. We verify removal using an exact two-sided binomial test rather than assuming the confidence target guarantees success.