“The problem with the traditional photofit images of crime suspects,” says Charlie Frowd, senior lecturer in psychology at the University of Central Lancashire, “is that they almost never work. You get a correct identification only about 20 percent of the time. That’s why we decided to take a completely new approach.” Frowd, 46, with colleagues in Newcastle and Stirling, developed EvoFIT, a software-based system that generates faces from witnesses’ sometimes vague recollections.
“In the old model, people select eyes, nose, mouth etc,” he says. “But you don’t think of a face in that way. You think in whole faces. EvoFIT breeds faces together.”
A crime witness is shown a screen of 18 faces and asked to choose two or three that are closest to that of the person they saw. These are morphed into a new selection and the process is repeated until they get a single image. They can tweak this, adjusting variables such as hairstyle and weight, and also what Frowd calls “holisitics” such as “pleasantness”, “honesty”, “threatening” and “health”. “We just thought,” he says, “what are the sensible things you might want to do to a face?”
The image can be published in print or in a slowly moving version, online or on television. “Animating the image brings out the individual,” says Frowd, “and it has doubled the recognition rate.” Frowd also encourages witnesses to look at still versions from the side. “You get a thinner image and the brain has to work out how to make it normal again. That’s often where the recognition takes place.”
The approach is paying off. EvoFIT is in use across nine police forces in the UK and Romania, and is being tested in Boston and Israel — delivering correct identification 74 percent of the time. Next up is 3D, says Frowd, “For when people haven’t seen a face straight-on.”