One of the worst features of Facebook is when a friend uploads a photo of you looking awful and then tags you in it so all of your friends can have a good laugh. You can untag yourself and burn your ugly outfit but the photo is still there. Worse still, your friend might have terrible privacy settings allowing anyone to see the picture.
Wouldn’t it be nice if you had more control over the privacy settings for photos you’re in? At CCS today Panagiotis Ilia spoke about a tool to give you just that. With their system, when a user uploads a photo it gets run through facial recognition software to identify the people in it. The photo is then split into layers so that each person gets control over the layer containing their own face. Users then choose who can see their face in the photo, with people outside the privileged group seeing a blurred face instead.
Ilia and his team ran experiments, showing people blurred face photos of random friends and found that 82% of those polled could not say which of their friends featured in the photos. Of those who could, their friend’s identity was given away in most cases by the other people in the photo, followed by their body or hair, and finally by their clothing, for example their characteristic Hawaiian shirt and running shoes.
A member of the audience asked what would happen if you used your favourite photo editing software to manipulate a person’s face so that the facial recognition software would be fooled but a human could still recognise the person, and was assured that should the facial recognition software fail to find a match for the person in the photo, they would appear blurred by default. This protects those who opt out of social networking from appearing in pictures. A potential issue here is if I get a selfie with my favourite celebrity then their face may be blurred by the software and no one will believe I really did meet Alan Hansen in a KFC.
If you make bad fashion choices, are regularly photographed in compromising positions, or just want to know more about the technical details you can read the paper here.