Differentially Private Imaging via Latent Space Manipulation
This addresses privacy concerns for individuals in social media and face recognition systems by providing a novel, privacy-preserving image obfuscation technique.
The paper tackles the problem of image privacy by introducing a method that manipulates latent spaces of a generative model to produce realistic facial images while satisfying local differential privacy, achieving formal privacy guarantees for individuals.
There is growing concern about image privacy due to the popularity of social media and photo devices, along with increasing use of face recognition systems. However, established image de-identification techniques are either too subject to re-identification, produce photos that are insufficiently realistic, or both. To tackle this, we present a novel approach for image obfuscation by manipulating latent spaces of an unconditionally trained generative model that is able to synthesize photo-realistic facial images of high resolution. This manipulation is done in a way that satisfies the formal privacy standard of local differential privacy. To our knowledge, this is the first approach to image privacy that satisfies $\varepsilon$-differential privacy \emph{for the person.}