Face search engines are posing a serious threat to privacy. But there are tools that can make your photos useless for fraudsters, stalkers, and Clearview.

While millions of people are daily posting millions of selfies on Instagram and Facebook, the army of those who want to erase their digital footprint from the internet also grows. People are getting more concerned about digital hygiene. Privacy laws, like GDPR and CCPA, are emerging around the globe. New services are offering users their help in managing their personal data on the internet.

On the opposite side of the barricade are the very controversial services like Clearview AI or PimEyes that can find your photos anywhere across the web in a matter of seconds. Including the photos you’re completely unaware of or would desperately like to forget about. And while Clearview AI is claimed to be cooperating only with law enforcement bodies, PimEyes is available literally to everyone: your next employer, your relatives, or all sorts of trolls, stalkers, fraudsters, and other “good” people of the internet.

So, is there a way to protect your photos from unwanted eyes? Let’s take a look at some solutions and check how they work.

Defacing

The developers from Playground created an iOS camera app that identifies people in the footage and blurs or blocks out faces or whole bodies.

Results of Anonymous cam app

The app can also distort voices in videos and removes metadata from files that can say where and when a photo or video was taken.

The developers of the app told The Verge that they were inspired by investigative journalists who needed a tool that would let them easily take records while keeping their sources anonymous.

If you want to completely impersonate a footage, this cam works great. And what if you still need to have an image of you on the internet, but want to stay invisible for facial recognition systems? There should be more elegant ways to withstand face search engines.

Refacing

There are also tools like CLEANIR and DeepPrivacy. They are based on different technologies. DeepPrivacy uses a Generative Adversarial Network (GAN). CLEANIR is based on a Variational Autoencoder (VAE). But they both do the same thing: they replace real faces with artificially generated ones. Both work with photos and videos. For technical details, you can read this great article comparing the solutions.

Unfortunately, there’s no easy way to play around with the tools, so here are some samples from the author of the article.

DeepPrivacy results

DeepPrivacy

Cleanir results

CLEANIR

Face swapping

Face swapping is a special case of refacing and another great way to anonymize photos. You take a photo of yourself, add a photo of another person, and get an image of an absolutely new human.

In Face Swapper by Icons8, you can play around with the degree of likeness between you and your avatar. If you seek for lesser resemblance, swap faces with someone who doesn’t look like you at all.

Face Swaper result with someone who doesn't look alike

If you want the result to look more like you, you can swap faces with your digital doppelgänger (Face Swapper has an integration with Anonymizer, see below).

Anonymizer result with digital doppelgänger

The developers of the app are sure that it has a great potential in creative and advertising. As for the privacy, they mentioned the following:

  • Anonymization of employees behind the helpdesk
  • Anonymization of employees in charge of sensitive matters (law enforcement officers, investigative journalists, etc.)
  • Medical companies can anonymize photos of patients
  • Dating services can offer more protection to their users
  • Platforms with adult content can attract more new models, if they can ensure their anonymity

For integration with Face Swapper companies can use API.

And here’s how Face Swapper handled the images from the CLEANIR and DeepSwap examples.

Face swapper vs cleanir and deepprivacyFface swapper vs cleanir and deepprivacy3

Doppelganging

Anonymizer by Generated Photos does not generate faces. Instead, it uses a huge gallery of 2.6M images that have already been generated by a GAN. It takes your photo, runs through the gallery, and returns a grid of your synthetic look-alikes where you can choose the one that you like the best.

Generated photos samples

These generated faces look absolutely realistic. They can hide your identity and yet provide the general idea of your appearance. You’ve probably already met some of them on Twitter, LinkedIn, or other social media.

Cloaking

Unlike all the previous solutions that visually modify original images, the privacy tool named Fawkes uses quite a different approach. The creators say that it makes some changes to a photo on the pixel-level. They call it “cloaking”. The changes are almost invisible to the human eye, but if you feed a cloaked photo to a face recognition system, it will fail to recognize the person in the photo. That’s what they say.

We tested the Windows version (1.0) of the tool and put the results through Google Image Search and PimEyes. Both said thank you, readily found similar images and identified the person.

Barak Obama_Google Image Search

Google Image Search

Barak Obama PimEyes

PimEyes

Wrapping up

Newton’s third law states that every force has a counter force. It’s nature. The same is applicable to technology. Computer viruses and antivirus software, geo-blocking and VPN, face recognition and face anonymization… They always come in pairs. Don’t panic, observe digital hygiene, keep calm and up-to-date.

Read also:




Subscribe to Icons8 Newsletter

Welcome to the community, buddy. We promise never to spam you.

pinterest linkedin VK reddit