How much of a face is a face: exploring reidentification potential with generative AI

Key Points

• The aim is to evaluate how generative AI affects the reidentification risk of obscured facial images.
• Analyzed the capabilities of generative AI in recognizing partially obscured faces.
• Utilized quantitative metrics to measure reidentification risk.
• Focused on the implications of AI technology on facial image privacy.
• Identified significant reidentification risks associated with generative AI.
• Quantitative metrics highlight the extent of risk for different types of obscured images.
• Results indicate that even partial obstruction can lead to high reidentification potential.