To see examples of the animated caricatures, click here

POLICE are to test out a system that transforms the Photofit faces of criminal suspects into animated caricatures up to seven times more likely to be recognised than standard likenesses.

The system starts with a witness’s recollection of a suspect’s face. It then “morphs” the image by exaggerating features remembered by the witness. Unlike a conventional static Photofit, the wide range of faces shown through the technique increases the chances of someone recognising who it is.

Developed by psychologists at the University of Central Lancashire and Stirling University, the technology will be piloted by Derbyshire police early next year. Peter Hancock, professor of psychology at Stirling University and a member of the research team, said the system would be ideal for a television appeal.

“A hunt to find the suspect in a serious case on a programme such as Crimewatch would be the ideal medium for us to animate a composite image and hence generate more leads for the police,” he said.

When faces were transformed into animated caricatures during the experiments, the scientists found they could be successfully matched to real people on 26% of occasions, even after a two-day delay. This compared with an average of just 3.7% for standard Photofits, in which a static computer-generated face is produced.

“This is a massive jump in the level of recognition and is really reliable,” said Charlie Frowd, a psychology lecturer at the University of Central Lancashire.

The system uses computer software to distort facial features and the way in which they are arranged. It does this by marking the different elements of the face – from its overall shape to the size of the nose and separation of the eyes – with 200 points.

Each point is then compared with the position of its counterpart on an average face of the same sex, age and racial group, produced by merging large numbers of different images. Any differences from the norm are then exaggerated to generate a caricature.

Different versions of the face are produced, ranging from an extreme caricature to a blander version with all the exaggerations removed.

Conventional static Photofits often include mistakes that can prompt the brain to reject a particular match. The animated caricature system overcomes this by including the blander version of the face, allowing poorly recalled features to be ignored. By morphing from the blander version to the extreme caricature, the brain is able to detect a possible match more often.

The animated caricature system was tested using images of footballers and other stars. Volunteers were shown the pictures and then two days later were interviewed to produce a Photofit image. People who would know the celebrities by sight were then asked to identify them from the pictures.

Vickie Burgin, the principal forensic services officer at Derbyshire police, said any tool that could improve detection rates would be explored. “Facial identification can be very effective and if we can put a name to a face it can greatly speed up an investigation,” she said.

Times Recommends

• Computer warriors pay for far-away mercenaries
• Divorce rate falls to 26-year low
• Boris adviser says Olympians tainted with blood