UCL in the News: Can 'facial recognition' help Madeleine McCann police analyse picture?

At best, all that a computer might conclude was that "this was a photo of a person who resembles Madeleine McCann a bit," Dr Simon Prince [UCL Computer Science].

The quality of the image was not even good enough to allow for a match - or what referred to as a "facial verification" - using the criteria that are typically applied in the analysis of, say, CCTV footage, Dr Prince said. …

"A typical digital camera might have 5 megapixels, meaning there are 5 million pixels in each image," Dr Prince said. "In this photo, the area covered by Madeleine's face makes up about 200 pixels, of which 10 are between her eyes, and that's not enough. In addition to the resolution being too low, the face is partially occluded by the shadow coming from the woman's arm, which further hampers identification."

Facial recognition technology was principally used to search vast databases of pictures - for instance CCTV footage or the internet - that would take humans far longer to examine, he said. …

"Facial recognition works well when it is comparing passport-style photographs where the subject is front-on, the lighting is consistent, and the facial expressions are similar," Dr Prince said.

"There can be problems, however, when the pose of the subject, as well as the lighting and expression, are different - that's where all current research in the area is focused."

Jonathan Richards, Times Online