Connect with us

Hi, what are you looking for?

Tech & Science

New software lets computers identify pixelated faces

Pixelation is used heavily across the media to obscure faces, vehicle license plates and restricted pieces of text. The technique generally works well, preventing people from identifying what’s beneath the blur without completely destroying the image with overlays or cut-outs. However, we may soon need to find an alternative to pixelation, according to researchers at the University of Texas.
WIRED reports that a team has developed software that is able to see through a pixelated mask to determine the identity of a person beneath it. The technology is able to process images as if the pixelation simply didn’t exist. It can identify images or text, such as that on license plates, and then send the information on for further processing.
The method is relatively mainstream and doesn’t require much specialist training. The team found that existing machine training techniques, such as using a set of sample data to program a computer, suffice. Even more advanced facial recognition methods could further increase the precision of the identification, defeating all existing methods of obscuring defined areas of an image. “The techniques we’re using in this paper are very standard in image recognition, which is a disturbing thought,” Vitaly Shmatikov, one of the research authors, said to WIRED.
To test their system, the researchers set out to defeat three privacy protection technologies. They began with YouTube’s proprietary blur tool which allows video uploaders to automatically blur out objects and faces in their content. The researchers’ computer could successfully identify faces pixelated by YouTube’s algorithm. It later named people featured in pixelated images created by Photoshop and the JPEG photo format’s Privacy Preserving Photo Sharing technology.
The rise of machine learning could make it harder than ever to maintain personal privacy online. Developing more advanced image obfuscation techniques may not be a viable solution. It’s possible that attackers would be able to continually bypass any kind of redaction mechanism, making it difficult for the media to use images of crime scenes or people speaking on condition of anonymity.
Shmatikov’s team hope their research paper will raise the standards for privacy technologies by demonstrating that no system is fully secure. For years, pixelation has been believed to be a simple and effective way to censor images. Its useful time may be nearly up though, thwarted by the increasing presence of advanced artificial intelligence and automated bots. “I hope the result of this paper will be that nobody will be able to publish a privacy technology and claim that it’s secure without going through this kind of analysis,” said Shmatikov.
The threat may be real but there are some limitations to the current version of the AI. Most notably, it is only able to identify things it’s been specifically trained to look for, rather than an entire image. For it to name a person pictured in a scene, it needs to have previously been shown a photo of the individual. An attacker could use online profile photos and social media uploads to train the system though, making practical application relatively straightforward.

Written By

You may also like:

World

Calling for urgent action is the international medical humanitarian organization Doctors Without Borders/Médecins Sans Frontières (MSF)

Business

The cathedral is on track to reopen on December 8 - Copyright AFP Ludovic MARINParis’s Notre-Dame Cathedral, ravaged by fire in 2019, is on...

Business

Saudi Aramco President & CEO Amin Nasser speaks during the CERAWeek oil summit in Houston, Texas - Copyright AFP Mark FelixPointing to the still...

Business

A recent article in the Wall Street Journal infers that some workers might be falling out of the job market altogether.