The Geo-Privacy Bonus of Popular Photo Enhancements

Abstract

Today's geo-location estimation approaches are able to infer the location of a target image using its visual content alone. These approaches typically exploit visual matching techniques, applied to a large collection of background images with known geo-locations. Users who are unaware that visual analysis and retrieval approaches can compromise their geo-privacy, unwittingly open themselves to risks of crime or other unintended consequences. This paper lays the groundwork for a new approach to geo-privacy of social images: Instead of requiring a change of user behavior, we start by investigating users' existing photo-sharing practices. We carry out a series of experiments using a large collection of social images to systematically analyze how photo editing practices impact the performance of geo-location estimation. We find that standard image enhancements, including filters and cropping, already serve as natural geo-privacy protectors. In our experiments, up to 19% of images whose location would otherwise be automatically predictable were unlocalizeable after enhancement. We conclude that it would be wrong to assume that geo-visual privacy is a lost cause in today's world of rapidly maturing machine learning. Instead, protecting users against the unwanted effects of pixel-based inference is a viable research field. A starting point is understanding the geo-privacy bonus of already established user behavior.

Image Privacy Prediction Using Deep Neural Networks

“…Location-based approaches. These approaches [Baokar 2016;Bilogrevic et al 2016;Choi et al 2017;Fisher et al 2012;Freudiger et al 2012;Friedland and Sommer 2010;Olejnik et al 2017;Ravichandran et al 2009;Shokri et al 2011;Toch 2014;Yuan et al 2017;Zhao et al 2014] leverage geo-tags, visual landmarks and other location contexts to control the location-based privacy disclosures. The geo-tags can be provided manually via social tagging or by adding location information automatically through the digital cameras or smart-phones having GPS.…”

Who's Afraid of Adversarial Queries?

“…A recent investigation concerned with the threat of cybercasing [4] examines the potential of image enhancements to block the inference of the geo-location of the photos that users take and post online. Our work differs from [4] in that we focus specifically on CBIR and we consider image queries that are explicitly designed to be adversarial. However, we adopt the same threat model, and we focus our investigation on a CBIR problem that is related to location because it involves images of buildings in cities.…”

Privacy and Audiovisual Content: Protecting Users as Big Multimedia Data Grows Bigger

“…Recent work demonstrates the potential of obfuscation to protect information about user location by foiling the content-based geo-location estimation methods discussed above. The study in [33] investigates the impact of common user image enhancements (filters and cropping) on the accuracy of geo-location estimation algorithms. Because the enhancements remove some information from images, the expected result is that filtered photos will be less geographically distinctive and more difficult to associate with the location at which they were taken.…”

“…It is important that users' sense of fun and feel for aesthetics are not disturbed by technology that protects their privacy. Recall that [33] investigated the ability of naturally occurring user practices of photo enhancement to reduce the accuracy of content-based geo-location estimation. Instead of developing a privacy-protection measure and hoping that users would adopt it, this work starts with a behavior that users already engage in and investigates its potential as the basis for a privacy-protecting technology.…”