Skip to content

Changing Faces

COMPUTER PROGRAMS have a hard time getting facial recognition right 100 percent of the time, and as it turns out, that task may be harder for humans than we think. According to ongoing research funded by the United Kingdom’s Economic and Social Research Council, researchers showed how easy it is for a person viewing photographs to wrongly believe that various photographs of the same person are actually of different people; the viewers were thrown off by just a change in expression in the photos.

Rob Jenkins, senior lecturer in psychology at the University of Glasgow and researcher on the project, says the work provides a new dimension to facial identification by pointing out that differences within faces are just as varied as differences between faces.

The findings have obvious security implications. When passport photos are taken, individuals are often prohibited from smiling in the photographs, because it was thought that smiles distort the face. Jenkins and the other researchers found that the opposite may be true, and frowns may actually be what make faces more difficult to identify, especially if that’s less likely to be how they look when at the checkpoint.

The reason blank faces on passports are not helpful, says Jenkins, is that people don’t tend to have blank faces in real life. They smile and have other expressions. “The smile can be a cue to identity. Remove the smile, and you remove that cue,” says Jenkins.

There isn’t a wide acknowledgement of how difficult facial recognition is, says Mike Burton, researcher on the project and psychology professor at the University of Aberdeen.

In their work, the researchers looked in part at what are known as the Bruce and Young face recognition units. They questioned how a representation could be built to recognize various incarnations of the same person but also rule out all other individuals, since a person’s face can look so different at different times. Even when the face is viewed straight on, as in the study’s photos, it takes only a change to the expression to make it more difficult for someone to determine if two photos of an unknown person are of the same person.

What makes the findings particularly significant for security is that humans are often called in when a computer program fails to assess whether a person matches the identification photo they are presenting. But Burton says photographs are just not a great mode of identification when a person is not familiar with the individual they are trying to verify: “Photos are an excellent means of identification if the viewer knows the person. If the person is unfamiliar, photos are very poor for identification. People are bad at using them (for example in photo-IDs), and automatic systems don’t seem to work either. Base-rates for other techniques, particularly iris-scans, are much higher.”

Burton says that there is a mismatch between research and social science policy in this area in that photographs are increasingly used for identification. He adds that problems with human ability to match people up have become obvious in the research into mistakes made in eyewitness identification.

Burton says that the ongoing work in the field of facial identification is important because “people just don’t know how hard face recognition is.” That’s partly the case because we think we see “just how
good face recognition is every day, as we recognize friends and family. However, we tend to generalize this level of accuracy to unfamiliar faces, and we shouldn't, because all the data shows that people are bad at unfamiliar face recognition.”