A Hertfordshire University researcher has proposed a privacy algorithm for social media that would automatically replace faces in photos with artificial ones, thus protecting people’s privacy online.
The algorithm, developed by Lily Meng and her colleagues, swaps the original faces in images and videos for completely artificial ones, whilst maintaining the original facial expressions. As a result, the new face looks perfectly natural but can't be associated with a specific real-life person.
“Companies like Google are already using face de-identification in their Street View images by blurring or pixelisation, but I think this approach is not sufficient,” said Meng. “For example, if the level of pixelisation is not high enough, human eye can still recognise the original face. If the level is high enough, than the original image is completely destroyed and you can’t even recognise there is a human face there.”
Meng envisions her algorithm could become an integral part of services such as Facebook to give people more control over who can see images of them, their children and friends on the Internet.
“I can control the photos I put on my Facebook account, but I can’t control my friends. They put photos with me on their Facebook account and even put a tag onto my face. I have no control over that,” she explained.
Ideally, the photos would be automatically run through the de-identification programme. Only those people the user wants would be able to access the original image, the rest would still see pretty images, only with non-existent but still natural-looking faces.
“The de-identified faces that we generate are synthesised, they are artificial, but they are created based on some real human faces,” Meng explained. “You can use the face of Shrek, or the Queen, for example, to create the new face, which would be impossible to be associated with a real human being.”
The only link to the original person would be that 'Shrek' or the 'Queen' would still have the original person’s lovely smile.
The system could also be used by researchers or criminal investigators who need to work with large quantities of videotaped interviews. Meng’s team, for example, has teamed up with a group from a university in the Czech Republic, which is developing a method to diagnose Parkinson’s disease from facial expressions.
“Parkinson’s disease affects facial muscles of sufferers and their facial expression becomes more rigid,” explained Meng. “We think we could diagnose the disease just by looking at the person’s facial movements. But in order to do that, we first need to analyse the facial expressions of a large group of patients and compare it with healthy people.”
With images and videos proving infamously easy to be hacked and leaked online, it is likely many people would object to having their videos taken, even if by scientific researchers, in relation to such a sensitive issue. People may have fewer objections if their faces were to be automatically replaced with that of Shrek.