DeepFakes for Privacy: Investigating the Effectiveness of State-of-the-Art Privacy-Enhancing Face Obfuscation Methods

DeepFakes for Privacy: Investigating the Effectiveness of State-of-the-Art Privacy-Enhancing Face Obfuscation Methods

연구 분야: Analysis

학회: AVI '22: Proceedings of the 2022 International Conference on Advanced Visual Interfaces

There are many contexts in which a person’s face needs to be obfuscated for privacy, such as in social media posts. We present a user-centered analysis of the effectiveness of DeepFakes for obfuscation using synthetically generated faces, and compare it with state-of-the-art obfuscation methods: blurring, masking, pixelating, and replacement with avatars. For this, we conducted an online survey (N=110) and found that DeepFake obfuscation is a viable alternative to state-of-the-art obfuscation methods; it is as effective as masking and avatar obfuscation in concealing the identities of individuals in photos. At the same time, DeepFakes blend well with surroundings and are as aesthetically pleasing as blurring and pixelating. We discuss how DeepFake obfuscation can enhance privacy protection without negatively impacting the photo’s aesthetics.