Perspectives on DeepFakes for Privacy: Comparing Perceptions of Photo Owners and Obfuscated Individuals towards DeepFake Versus Traditional Privacy-Enhancing Obfuscation

Obfuscating people’s faces using synthetically generated faces, i.e., DeepFakes, has been shown to be effective at privacy preservation. While recent work showed that DeepFake obfuscation is well perceived by viewers, the perspectives of a) the owner of the obfuscated photo, and b) the person that is being obfuscated, remain unclear. This paper reports on the results of a user study where participants uploaded their own group photos, in which they appear, and applied obfuscation techniques to both themselves and others in the image. The obfuscation methods included DeepFakes and four traditional techniques: blurring, pixelating, masking, and avatars. Our findings show that both photo owners and obfuscated individuals perceive DeepFake obfuscation as significantly more effective in protecting privacy compared to the traditional methods, and was found to integrate well with the environment.