DeepFake How do you break a face swap scam? Make him turn sideways

A user on Hacker News also quipped.

I recently went to an unknown bank to get a card, and they needed a side view of me, which I was confused about at the time, but now I know why.

You can also identify a fake face by waving your hand in front of your face

There is a small way to tell if a person is a DeepFake fake on a video call, other than by turning his face sideways: by waving your hand in front of you.

If it is a fake face, the superimposition of the hand on the face image may be misplaced and there will be a delay in the hand shaking process.

In the case of the face-swapping widow and Professor X, the hand and face superimposition occurs incorrectly

▲ Hand and face superimposed incorrectly when swapping faces between Widow and Professor X

Real-time DeepFake has the problem of superimposing realistic occlusions over unrealistic facial images, an operation commonly referred to as "masking" or "background removal".

And the real-time DeepFake model needs to be able to perform keying at will, on demand, to a convincing level.

However, there are often many obfuscating masks that can interfere with the 'keying' process, for example, masks with facial features that can cause 'distress' to the model and make the 'keying' process It is difficult to do so.

The rapid movement of an occluder can make 'keying' very difficult by waving your hand in front of the faked face, causing significant delays and affecting the quality of the overlay.

One More Thing

Face-swapping crimes are not far-fetched, and there have been media reports of suspects using DeepFake face-swapping to conduct remote interviews for IT jobs in an attempt to hack into companies and gain access to their customer or financial data, as well as corporate IT data and professional information.

The Federal Bureau of Investigation (FBI) has written to its Internet Crime Complaint Centre about receiving multiple complaints of people using stolen information and deeply falsified videos and voices to apply for remote technology jobs.

In the cases described in the federal agency's May report, some face-swapping suspects operated through several layers of shell companies, which made it more difficult to identify them.