Detecting faked photos and videos quickly and easily

"Keeping pace with increasingly sophisticated hoaxers is a huge challenge.”

Anthony Sahakian, CEO of Quantum Integrity SA

A video in which Barack Obama berated his successor, Donald Trump, caused a huge stir. Only it wasn’t actually Obama; the words were spoken by an actor. Deepfake videos are now so realistic it is almost impossible to recognise them as fake with the naked eye. “It still takes a huge amount of effort and expertise to produce high-quality deepfake today,” pointed out Anthony Sahakian. “But that’s all soon set to change.” Nevertheless, he explains that the methods used by hoaxers are becoming increasingly sophisticated, and the consequences more and more devastating: for instance, faked photos and videos are being used in insurance fraud, showing damage to cars that does not really exist or fictitious goods that are reported stolen.

Sahakian’s company Quantum Integrity already provides software that can detect fake images. Working with EPFL’s Multimedia Signal Processing Group MMSPG, Quantum Integrity plans to upgrade its deepfake detector so that it can be deployed more widely. The aim is to set up a website where videos can be uploaded for review. “We couldn’t develop this system alone, because the technical issues involved are extremely complex and costly,” says Sahakian. “Prof. Touradj Ebrahimi and his team are the best support we could possibly hope for.” The Swiss start-up is especially reliant on this support, given that competition in this field is fierce: both Facebook and Microsoft plan to invest 10 million US dollars in deepfake recognition software over the next few years.