VFRAME: Visual Defense Tools
Use computer-vision to shield privacy in video
Visible data shares many of the same risks as wireless data yet visual privacy is often overlooked in the field of information security studies as separate and less relevant. As computer vision becomes increasingly adept at understanding the visual domain, differences between existing protocols for processing wireless data and emerging protocols for processing visible data (computer vision) become less apparent. Ultimately, images and video are wireless data too, and they are exposed to an increasing number of attacks on visual information privacy with less technologies for protection. Visual Defense Tools will explore and prototype computer vision methods for visual privacy through visual obfuscation and minimization techniques, mostly related to biometrics. The goal will be to build a conceptual road map and functional open-source prototypes to stimulate future development of more accessible visual privacy technologies.
- The project's own website: https://vframe.io
Why does this actually matter to end users?
We live in an time where it seems there are video camera's everywhere. Not just on a payment terminal, a traffic sign or at the entrance of an office building either. Every day hundreds of millions of people take their phone from their pocket and start filming themselves and others. And they do not just record it for their personal use, they tend to put it online. But in many instances you are not the only one captured on screen. The person next to you might not want to be put online, for whatever reason. They might be underage. They might have their credit card out. In the "selfie" mindset, those concerns may not matter much. After all, what can they do? All they want is to capture the moment, and you just happened to be in the wrong place at the wrong time.
Now combine this with the ever increasing power of computers to process video. Todays capabilities go far beyond face recognition in pictures. Computers can now almost realtime process video recordings and live streams. Connect that in your mind with the insane amount of information about people already available online, for instance through social media. This could end up as the 'coup de grace' for privacy in the public sphere: point your camera to a girl in a bar, and their name, income, hobbies and past sex life automatically pop up. The creep factor of this is enormous, and so is the potential for abuse. That is probably the reason why some of the largest privacy violators on the planet have invested literally billions of dollars in building technical capabilities to that effect.
All of a sudden, you end up with a spillover from the digital world into the real world. If we want to protect privacy in the public sphere, we will need to create technologies that can counter this trend of crowd-sourced public surveillance. It is probably infeasible and not desirable to ban camera's. There are many beneficial uses for them too, after all. So the next best thing is to create technologies that can protect the privacy of people that would be captured on video against their will. We can use video recognition technology the reverse way: wipe out everything concerning the people you did not explicitly get permission from. It doesn't really matter to your vlog viewers if you cannot see the faces of the people walking behind you - or randomise their clothes. All we need it the technology to do this automatically. This is the goal of the VFRAME project: it aims to build a first, exploratory tool as a first step to restore visual privacy. In the end, this is probablyu something that should be turned into legislation so that all consumer devices behave like this by default. So that next time that first kiss on a bench in the park, will not be part of internet history.
This project was funded through the NGI0 PET Fund, a fund established by NLnet with financial support from the European Commission's Next Generation Internet programme, under the aegis of DG Communications Networks, Content and Technology under grant agreement No 825310.
