Science

Morphing helps criminals across the border

| Jelle Posthuma

Digitally altering your face is all the rage. Think of those popular ageing apps that you can use to find out what you will look like when you are old. Did you know that criminals use so-called face morph apps to illegally travel across the border? Luuk Spreeuwers, associate professsor in the Data Science department, is researching this computer technology.

‘Let me show you something first,’ Spreeuwers says as he takes out his phone. ‘This is a face morph application.’ On the screen of Spreeuwers’ iPhone are two passport photos of random people. At the press of a button, one image is superimposed over the other, creating an entirely new, completely unique face. ‘That’s how easy the morphing process is. Myriad apps like this one are available for free.’

Criminals

The application seems harmless enough at first glance. It looks like something you would try out with your friends for a laugh. However, the app has proven very useful to criminals. Here's why: suppose a criminal knows the law is closing in on them and they want to leave the country. However, they have no chance of making it through customs without getting caught. Using someone else's passport to be able to travel out of the country would be the perfect solution. 

That is where the face morphing process comes in, Spreeuwers explains. ‘First of all, the criminal has to find someone who looks like them. It can be a random person or a family member, as long as there is some physical resemblance. The criminal approaches them and asks them if they would like to make some extra cash on the side. If they say no, there is always the threat of violence to persuade them. The criminal then asks the person in question for a passport photo. The culprit's picture and the criminal's own are fed into one of these easy-to-use apps to create a morph.’

It is now time for step two. ‘The culprit is given a print-out of the manipulated picture and goes down to town hall to apply for a new passport. The official behind the desk is tasked with making sure the passport photo matches the person standing in front of them. They are used to imperfect matches, because people tend to change over time. Someone might decide to grow a beard or get a new pair of glasses. The clerk cannot tell the difference between the fake image and the person at their desk. The culprit therefore has no trouble getting a new, valid and government-issued passport. They hand it over to the criminal, who looks enough like the morphed picture to make it through customs without getting caught. Thanks to their new identity, they have managed to leave the country after all.’

'Human observers are hardly able to recognise a good morph'

Facial characteristics

How is it possible that the fake passport arouses no suspicion at town hall or during the manual and automated customs inspections? ‘Human observers are hardly able to recognise a good morph,’ Spreeuwers explains. ‘The same goes for the automated passport check at airports. Various studies have shown that morphs can pass any standard automated passport check. That is a scary thought, of course.’

‘As I mentioned, people are easily fooled by morphs. In general, only the centre portion of a face is used. Everything around it, such as a person's hair and the background, are not usable. If you combine a thick head of hair and a bald pate, the effect looks very weird. We call that “ghosting”. Here's what people do instead: they only use the centre portion of the face and paste that onto the face of someone else. If you do it well, it becomes almost impossible for humans to recognise that they are being deceived.’

Automated facial recognition technology also has trouble identifying morphs, Spreeuwers explains. ‘During an inspection, this technology tries to detect facial characteristics. The technology is trained by being fed a vast quantity of data consisting of millions of passport photos. When you show a ton of faces to a self-learning system, it will extract the main characteristics of a face on its own.’

‘A good morph contains characteristics that match two different faces: there is a degree of overlap. That is exactly the problem. As a result of this overlap, the facial recognition technology cannot tell the fake image from a ‘real’ face, because it looks a bit like both of the original faces that were used to create the morph. All it takes is a moderate degree of resemblance between the two individuals. Even an automated facial recognition system will let the criminal with their false passport through.’

'The problem might be much bigger than we think'

Natural or unnatural

The Netherlands and Germany have already called the attention of the European Commission (EC) to the topic of morphing. A research proposal entitled ‘State of the Art of Morphing Detection’ (SOTAMD) has been submitted and approved, Spreeuwers says. ‘We want to generate a database filled with highly challenging morphs. In a way, we are putting ourselves in the criminals’ shoes. That is how the research team wants to develop a benchmark that poses a challenge to the morphing detection algorithms and can train them.’

‘We want to find out if morphs produce any unnatural distortions in faces. Are morphs fundamentally different from real pictures, or are they realistic images of non-existent individuals? In other words, does combining two pictures create a new person or is the result something unnatural? If any unnatural distortion occurs, that can be detected.’

‘The scope of this problem is not entirely clear yet. There is one documented case of a criminal who made it across the border with a morphed passport photo. In some other cases, journalists have tried to do the same. This tells us it is not impossible. Since detection is such a challenge, the magnitude of the situation is hard to guess at. The problem might be much bigger than we think. In any case, it is a serious threat.’