快猫短视频

IMAGES made by merging several photos of the same face are easier for computer software to recognise than ordinary snapshots. This finding suggests that using 鈥渁veraged鈥� faces on passports might help security systems make better matches of people against criminal watch lists.

Psychologists Rob Jenkins and Mike Burton of the University of Glasgow in the UK have found that people are better at recognising faces from photos that have been averaged than from individual photos. They suggest that this is because we see a person from different angles and in varied lighting, so the image we store in our brains 鈥� and later use to recognise them 鈥� is effectively an average of those experiences.

鈥淭he facial images we store in our brains may be an average of many views鈥�

To see if averaging might also improve facial recognition software, the researchers turned to the site . Powered by a facial recognition program called , one of the best available according to a 2006 test by the US National Institute of Standards and Technology (NIST), the site has a feature that finds out which face in its database of famous people a user most resembles. Jenkins and Burton used it to test their averaging hypothesis.

They fed in 20 photos each of 25 different male celebrities, including Bill Clinton and Jack Nicholson. MyHeritage correctly identified the celebrity only 54 per cent of the time. Then the researchers made a composite photo for each of them, which they built by combining all the photos of that celebrity used in the first round. This time MyHeritage identified them every time (Science, ).

The pair also built averaged faces that excluded those photos that had a correct match in the first test, in case they were skewing the results. MyHeritage matched them correctly 80 per cent of the time.

To merge the faces, the researchers identified landmarks on multiple images of the same person, such as the corners of the eyes or the mouth. Then they fed the images into software which used the landmarks to align the photos. If the photos show a person at different angles, the software stretches the image so that landmarks on different pictures are aligned. Then the software merges the photos. 鈥淭he good thing about averaging is that it washes out the things like lighting and pose that differ between photos,鈥� Jenkins says. 鈥淵ou are extracting the essence of a person鈥檚 appearance.鈥�

Jonathon Phillips at NIST says that the approach is original, but needs to be tested on a larger database before being used in airports.