Technology

Computers could identify people by their ears

Computers could identify people by their ears
Southampton's image ray transform is able to locate and extract ears in images of peoples' heads
Southampton's image ray transform is able to locate and extract ears in images of peoples' heads
View 1 Image
Southampton's image ray transform is able to locate and extract ears in images of peoples' heads
1/1
Southampton's image ray transform is able to locate and extract ears in images of peoples' heads

If you’ve watched any spy movies, then you’ll know that biometric security systems can recognize individuals based on physiological traits such as their fingerprints, handprints, faces and irises. Well, you may soon be able to add “ears” to that that list. Scientists from the University of Southampton’s School of Electronics and Computer Science have used a program called image ray transform to achieve a 99.6 percent success rate in automatically locating and isolating ears in 252 photos of peoples’ heads.

According to Southampton’s Prof. Mark Nixon, ears are a good biometric indicator. Their unique structure doesn’t change as the person gets older, they aren’t affected by facial expressions, and they are always predictably displayed against the side of the head – complete faces, by contrast, can end up with all sorts of chaotic backgrounds behind them, making things more difficult for computer imaging systems.

The image ray transform used in this study utilizes a “pixel based ray tracing technique” and a subset of the laws of optics, analyzing the way that light reflects off of objects in pictures. It is able to identify and extract tubular and circular features from images, such as the helix (the curved outer rim) of someone’s ear. The system then creates an isolated image of just the ear, even allowing for hair or spectacle arms covering part of it. The ear’s owner could then be identified by matching that image to one in a database of ear images.

The research was detailed in the paper A Novel Ray Analogy for Enrolment of Ear Biometrics which was recently presented at the IEEE Fourth International Conference on Biometrics: Theory, Applications and Systems.

2 comments
2 comments
windykites
Hopefully, the person being identified will conveniently turn sideways, otherwise this will not work. I thought that facial recognition was fairly accurate these days? Also there would have to be a data base of ear images in the first place.
Matt Rings
CSI:New York had an episode that solved a murder case by his \"ear-print\" on a car window...