Where to be wary: The impact of widespread photo-taking and image enhancement practices on users' geo-privacy

TL;DR

This paper investigates how common photo-taking and editing practices can unintentionally protect users' geo-privacy by hindering geo-location estimation, suggesting a new approach that leverages existing user behaviors.

Contribution

It demonstrates that widespread user practices can naturally protect geo-privacy and provides a systematic analysis of how these practices impact geo-location inference.

Findings

Widespread photo practices can protect over 50% of images from geo-location.

Existing user behaviors can be harnessed for geo-privacy protection.

Geo-privacy protection can be achieved without changing user habits.

Abstract

Today's geo-location estimation approaches are able to infer the location of a target image using its visual content alone. These approaches exploit visual matching techniques, applied to a large collection of background images with known geo-locations. Users who are unaware that visual retrieval approaches can compromise their geo-privacy, unwittingly open themselves to risks of crime or other unintended consequences. Private photo sharing is not able to protect users effectively, since its inconvenience is a barrier to consistent use, and photos can still fall into the wrong hands if they are re-shared. This paper lays the groundwork for a new approach to geo-privacy of social images: Instead of requiring a complete change of user behavior, we investigate the protection potential latent in users existing practices. We carry out a series of retrieval experiments using a large collection of social images (8.5M) to systematically analyze where users should be wary, and how both photo taking and editing practices impact the performance of geo-location estimation. We find that practices that are currently widespread are already sufficient to protect single-handedly the geo-location ('geo-cloak') up to more than 50% of images whose location would otherwise be automatically predictable. Our conclusion is that protecting users against the unwanted effects of visual retrieval is a viable research field, and should take as its starting point existing user practices.