Scholarlink

<p>The recent spread of pervasive technologies like Mixed Reality (MR), which have cameras continuously collecting visual data from spaces around us, raises serious privacy concerns. While direct sensitive content appearing in a feed seems like the only problem, sensitive information might leak indirectly from other subtle visual cues, particularly shadows. Existing privacy-preserving frameworks are not tailored for MR environments where private objects are often central within a dynamic three-dimensional (3D) space, and neglect shadow privacy leakage. To address shadow leakage, we propose integrating a shadow obfuscation algorithm into our 3D visual privacy-preserving framework, which adjusts the obfuscation region during non-trivial motion and uses machine learning to predict and prevent obfuscation failures in real time. Our proposed system improves state-of-the-art with 25% fewer privacy failures and a 32% stronger privacy-utility trade-off. Such improvement could be raised to 69% fewer failures and a 3.3x higher tradeoff when ideal shadow detection is reached.</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

No shadow left behind: diminishing shadows for enhancing privacy guarantees in mixed reality

  • 期刊论文

  • 发表日期：

    2025年10月9日

查看全文
  • Salam Tabet,
  • Kareem Bouakl,
  • Ayman Kayssi,
  • Imad H. Elhajj

摘要

The recent spread of pervasive technologies like Mixed Reality (MR), which have cameras continuously collecting visual data from spaces around us, raises serious privacy concerns. While direct sensitive content appearing in a feed seems like the only problem, sensitive information might leak indirectly from other subtle visual cues, particularly shadows. Existing privacy-preserving frameworks are not tailored for MR environments where private objects are often central within a dynamic three-dimensional (3D) space, and neglect shadow privacy leakage. To address shadow leakage, we propose integrating a shadow obfuscation algorithm into our 3D visual privacy-preserving framework, which adjusts the obfuscation region during non-trivial motion and uses machine learning to predict and prevent obfuscation failures in real time. Our proposed system improves state-of-the-art with 25% fewer privacy failures and a 32% stronger privacy-utility trade-off. Such improvement could be raised to 69% fewer failures and a 3.3x higher tradeoff when ideal shadow detection is reached.