News Archive

Stealing identities and falsifying security records are social problems of the 21st century that can result in crime, immigration fraud and more.

But the answer to the problem may be very close - right on the side of your head.

A team of computer scientists has come up with compelling results for detection and identification using the human ear.

Dr Syed Islam and his PhD supervisors Winthrop Professor Mohammed Bennamoun, Professor Robyn Owens and Dr Rowan Davies from the School of Computer Science and Software Engineering have recently demonstrated accurate recognition results for up to 99.9 per cent of detection tasks and 95.4 per cent of identification tasks, using two- and three-dimensional ear images.

Dr Islam was awarded a Distinction for his PhD on ear biometrics - biometrics is the application of statistical methods to the measurements of biological objects.

He said that traditional recognition systems based on identity cards and passwords could be easily be lost, damaged or faked.

"We propose to use ear images, which are common to all, as distinctly different for every person (even in twins) as fingerprints, and do not significantly change between the ages of about eight and 70 years," Dr Islam said.

Even hair, earrings and earphones cannot obscure ear identification

"Ear shapes also do not change when facial expression changes and they can be easily captured non-invasively, even in public places."

He said the team had developed a fast technique which could detect an ear within 7.7 milliseconds, from a large two dimensional side face image.

"Corresponding 3D data is extracted to compute a final match in only 2.28 seconds on a standard PC," he said. "The approach works well even when the ear is significantly covered with hairs, earrings or ear-plugs.

"Proposed complete, fully-automatic, highly efficient and accurate techniques will make significant improvements in thwarting identity frauds. They will help to create trust in e-banking, assist immigration and passport control, and reduce the use of plastic cards, with the added bonus of safeguarding the environment."

Dr Islam said the non-invasive quality of ear biometrics would appeal to people who needed a fast, simple means of identification.

"To use a fingerprint, you must put your finger on something; to use the iris as identification, you must look closely into a camera; a signature obviously requires you to sign something; and you need to speak to be able to use voice recognition techniques.

"You need only walk through something like a doorway for a biometric image of your ear to be captured," he said. "You can't fake it without having your ear cut off and an artificial ear put on your head in its place."

Dr Islam's work took Associate Professor Ajmal Mian and Professor Bennamoun's work on face recognition a step further. "At first, we didn't realise that ears were all so distinctly different," he said.  "When we did, we became very excited."

A person's ears are 90 to 95 per identification. The team concentrated on using left ears, which provided successful results even when covered up to 50 per cent by hair, earrings or earphones. "We only need to match key local features, not the whole ear," he said.cent symmetrical and identical, so either ear could potentially be used for

"But it would be fairly rare for people younger than eight or older than 70 to partake in illegal activity, so, despite those early- and late-life changes, the ear still provides a reliable security check."

The research findings are available online in the International Journal of Computer Vision. Dr Islam was awarded the UWA Early Career Postdoc Special Commendation Award 2011 for this publication.

Dr Islam is currently working as a Research Assistant Professor in the School of Dentistry, extending his biometric work for the improvement of orthodontic surgery, especially for the management of obstructive sleep apnoea. He is funded by four competitive grants.

Published in UWA News , 5 March 2012