Publishing Video Data with Indistinguishable Objects.

Area of Science:

• Computer Science
• Information Security
• Multimedia Systems

Background:

• Video data sharing offers community benefits but poses privacy risks.
• Existing privacy methods like blurring lack quantifiable risk assessment and background knowledge protection.
• Current techniques fail to address inferences drawn from sensitive objects in videos.

Purpose of the Study:

• To define a novel privacy notion, epsilon-Object Indistinguishability, for sensitive objects in videos.
• To propose VERRO, a video sanitization technique for privacy preservation.
• To generate utility-driven synthetic videos with indistinguishable objects.

Main Methods:

• Defined epsilon-Object Indistinguishability for sensitive objects (humans, vehicles).
• Developed VERRO, a video sanitization technique using random generation of synthetic videos.
• Ensured object indistinguishability within generated synthetic videos.

Main Results:

• VERRO generates utility-driven synthetic videos with rigorous privacy guarantees.
• Experimental results show synthetic videos retain high utility compared to originals.
• The proposed method effectively protects sensitive objects from privacy risks.

Conclusions:

• VERRO provides a robust solution for privacy-preserving video sharing.
• The epsilon-Object Indistinguishability framework offers quantifiable privacy.
• Synthetic videos generated by VERRO can be safely disclosed to untrusted recipients.