We are in the UK. The investigation into an armed robbery has been at a standstill for a long time. The only things the police have to go on is some poor-quality video footage and a few tentative leads. However, after 18 months without a breakthrough, the case takes a sudden turn: by pure coincidence, one of the investigators notices a face in the grainy CCTV footage from a busy shopping street – and the perpetrator is promptly arrested.

Another UK police officer is said to have arrested a suspect in a similar case, eight months after seeing a composite sketch of the individual.

A third police officer reports having made an arrest after recognizing a scar from 10 years earlier.

These three investigators belong to an exclusive group of people: the super-recognizers.

“Super-recognizers are not only exceptionally skilled at comparing and remembering faces, they can also identify individuals under conditions that normally make facial recognition very challenging. Such as in unclear and dark images or in environments filled with distracting elements,” explained professor Christoph Busch, head of the Norwegian Biometrics Laboratory at the Norwegian University of Science and Technology (NTNU) in Gjøvik.

We will soon see that the research he has recently conducted, in collaboration with colleagues from the UK, Germany and Norway, suggests that the abilities of super-recognizers extend even further than previously thought. Firstly, however, a few words about memory.

Generally speaking, human memory is not particularly reliable. One of the world’s most famous memory researchers, Elizabeth F. Loftus, states that remembering has more in common with putting together a puzzle than playing back a video recording. Sometimes, all it takes is a slightly leading question for the pieces of information about what actually happened to get mixed up with new information, resulting in false memories.

As a result, incorrect eyewitness accounts have led to many miscarriages of justice over the years. According to statistics from the non-profit organization the Innocence Project, over 60 per cent of the wrongfully convicted individuals they have helped to exonerate were convicted due to false eyewitness testimonies.

This knowledge highlights just how impressive the super-recognizers’ ability to accurately compare and remember faces is.

“It really is like having a superpower,” said Busch.

For a long time, facial recognition research focused primarily on people at the opposite end of the scale: people who, for various reasons, are no longer able to recognize even the faces of their loved ones. It is estimated that approximately 2 per cent of the population is affected by this condition.

When it became known to researchers about 15 years ago that a similar proportion of people are super-recognizers, it did not take long before the police began exploring the possibilities of making use of this special ability.

The first example of this was when London’s Metropolitan Police, affectionately known as ‘The Met’, brought in 20 super-recognizers to identify individuals who had stolen items or committed vandalism during the 2011 riots in London. These super-recognizers were recruited from within their own ranks, and managed to identify over 600 perpetrators by studying video footage from surveillance cameras and comparing it to archived mugshots. Over 70 per cent of the individuals identified ended up being convicted after physical evidence and confessions were obtained.

This success led the London police to quickly establish a permanent special unit consisting of super-recognizers. According to their own statistics, this significantly increased the annual number of identifications. It did not take long before police in other countries followed suit, and super-recognizers are now being used in Germany, Australia, Austria, the United States and Switzerland.

Exactly how far the abilities of super-recognizers extend is still an unanswered question. However, research conducted by Busch and his colleagues now shows that the scope of this very special ability may be greater than previously assumed.

“We have been conducting research for many years on how to detect whether images have been manipulated, for example in connection with border control. Unlike in Norway, in many European countries, people can provide their own photos when obtaining a passport. This has made image manipulation a widespread problem,” Busch said.

While some forms of manipulation are relatively harmless, others are far more problematic.

“Some people just want to look their best in their passport photo. They might use an app to raise their eyebrows or smooth out some wrinkles. Other people, however, use manipulation to sneak through border controls,” he said.

One of the most cunning methods to achieve this is ‘morphing’.

“Morphing is when you blend two photos of two different people into one. The photo will then share similarities with both of the original faces, making it possible for two people to use the same passport,” he said.

It has been known for quite some time that morphing can cause big problems. But it was not until a 2022 study showed that professional border control officers accepted nearly one-third of the morphed images they were presented with that serious alarm bells started to ring.

This is part of the reason why researchers have now investigated whether the abilities of super-recognizers can be transferred to other tasks. These are tasks that are not strictly related to recognition.

“Not only are the super-recognizers we studied significantly better than the control group at detecting morphed images, they are also better at recognizing when they are shown authentic, unmanipulated images,” Busch said.

The fact that super-recognizers perform so well on other tasks that humans usually struggle with is good news. Not that this means all positions involving this type of task can now be filled with super-recognizers.

Frøy Løvåsdal is a senior adviser at the National Police Directorate and has been collaborating with the biometrics team at NTNU in Gjøvik since 2012. She is currently pursuing a public doctoral degree at the same department as Busch and has been involved in the aforementioned research. Løvåsdal explains that perhaps the most promising use of super-recognizers for the Norwegian police is as a kind of mentor.

“The really good morphed images are almost impossible to detect, both for humans and for the algorithms we currently have. Fortunately, it looks like training helps,” said Løvåsdal.

A lot suggests that the super-recognizers’ very special abilities are innate – but that does not mean it is impossible to train other people to become better at recognizing faces and manipulated images.

“You can’t just decide to become a super-recognizer. However, our research shows that participants’ ability to detect manipulated images improved simply by taking part in the experiments,” explained Løvåsdal. She specifically points to two areas where super-recognizers can be useful.

“If we manage to understand what makes super-recognizers so good at remembering and comparing faces, we can use that knowledge to develop better training programmes for the relevant jobs.”

By equipping super-recognizers with sensors that track their eye movements, researchers have already gained some understanding of which areas they focus on most when looking at a person’s face. Exactly what they focus on within these areas, however, is still unknown. If they manage to solve this puzzle, Løvåsdal believes that super-recognizers may also help us develop better algorithms for automated facial recognition.

Even in a world with better algorithms, it is still important that humans are part of the process.

“Our research suggests that a combination of algorithms and human recognizers is probably the best solution. While the algorithms can handle many of the tasks, humans need to step in when uncertainty arises. It is generally a good idea to have humans involved in making important decisions when needed,” said Busch.

Løvåsdal agrees, stressing that neither algorithms nor super-recognizers will replace any of the people already working on these tasks in the police.

“As mentioned, we are primarily interested in what super-recognizers can teach us about improving the training programmes for our employees. There may be enormous potential here. At some point, we will probably also need to consider whether to test for this ability in future appointment processes for relevant positions,” she said.