The psychology of human face recognition

The ability of humans to recognise faces tends towards two extremes. On the one hand we are extremely good at recognising faces we are familiar with and can do so despite natural changes in expression, viewpoint and hairstyle as well as image manipulations such as inversion. However, when it comes to faces that we have been exposed to only briefly things are very different. Eyewitnesses to a crime find it very hard to describe and particularly to identify the perpetrator and are very susceptible to exposure to similar faces and the instructions given to them. The use of eyewitness evidence can also prove problematic as the witness may well express an inappropriate level of confidence in their decision. Of course, an eyewitness is often reliant upon their memory for a face they were briefly exposed to several weeks ago, but things do not appear to improve when the task becomes a purely 'perceptual' one involving no memorial load. Research has found that 'operators' perform poorly when they are required to match two facial images, as is often the case in identification from surveillance footage or the use of photo-identity cards. The present paper examines the face perception abilities of humans in relation to three separate forensic scenarios, photo-credit cards, identification parades and facial composite construction and looks at whether these form a suitable benchmark for machine recognition.