Posted: 12 Nov 2010 01:32 PM PST
On a planet hosting 6.7 billion human beings, having proof you're unique is of tantamount importance. The ear, it turns out, may be the best identification yet.
Through a new shape-finding algorithm called "image ray transform," which boasts 99.6 percent accuracy, according to a study presented at the IEEE Fourth International Conference on Biometrics Sept. 29, the outer ear may prove to be one of the most accurate and least intrusive ways to identify people.
Fingerprint databases of U.S. government agencies alone store the records of more than 100 million people, but prints can rub off or callous over during hard or repetitive labor. With the advent of computer vision, researchers and identification industries are seeking easier and more robust biometrics to get their hands on.
"When you're born your ear is fully formed. The lobe descends a little, but overall it stays the same. It's a great way to identify people," said Mark Nixon, a computer scientist at the University of Southampton. and leader of the research.
"There's real power in using the appearance of an ear for computer recognition, compared to facial recognition. It's roughly equivalent if not better," said computer scientist Kevin Bowyer of Notre Dame, who is pursuing his own ear-recognition technology and not involved with Nixon's work. "If you've got a profile image for someone, this is a great way to use it."
Recent technologies use computer vision to convert human features, such as faces and irises, even the gait of a person's walk, into reliable alternatives to fingerprints. Nixon and his team have pursued using ears as one biometric for years, and through what he called a "blue-sky research effort," his colleagues created the highly capable image-ray-transform algorithm.
The technology can identify an ear time after time with 99.6 percent accuracy. It works by unleashing a ray-producing algorithm on an image to seek out curved features. When a ray finds one, the software draws over the part and repeats the analysis. In a few hundred or thousand cycles, it cleanly paints the ear more than any other face structure.
"The rays fly around the image and get caught in tubular things. The helix, or outer edge, of an ear is a wonderful tube that rays keep hitting," said Alastair Cummings, the Southampton University computer scientist who developed the algorithm. "There are dozens of ways of doing ear biometrics, but this is a very good one."
From there, another program turns the curves into a unique set of numbers, something that could be used as an ear-based ID.
Nixon and Cummings acknowledged some limitations of the system, including hair covering the ears, less-than-ideal lighting conditions, and different IDs generated from different angles. And using the ear as a biometric isn't without critics.
"I have seen no scientific proof that the ear doesn't change significantly over time. People tend to believe notions like these, and they are repeated over time," said Anil Jain, a computer scientist at Michigan State University who was not involved in the study. "Fingerprinting has a history of 100 years showing that it works, unless you destroy your fingerprints or work in an industry that gives you calluses."
Using the ear is not about replacing existing biometrics such as fingerprints, Bowyer said. Rather, it's about supplementing them, especially when it comes to catching crooks.
"It's easy to say, 'Hey there's fingerprints, faces and irises, why do we need more?' For some applications that's a valid question," he said. "But when you're doing surveillance, where a person isn't being cooperative for obvious reasons, you want anything you can get. If you have images of ears, it's dumb to throw that away."
What's more, he says, there really aren't studies proving the agelessness of any human biometric — including fingerprints.
"Who over the age of 40 could think these things don't age?" Bowyer joked. "Some have said 'irises are for life,' but in some of our lab's work we've noticed degraded biometric performance even in those."
To address limitations of the approach, the team is looking to demonstrate that ears do hold up over time. In addition, the researchers hope to pair their new biometric with other computer-vision technologies, such as face recognition, to bolster its reliability. And if the algorithm can be made to work quickly in three dimensions, a fuzzy clip of a criminal walking by a security camera could be turned into grade-A courtroom evidence.
"We've shown we can use ears, but can we process data that comes from a sort of normal scenario? That's the real challenge," Nixon said.
Images: Alastair Cummings/Southampton University (demo) 1) A man's profile processed by the image-ray-transform algorithm, with a multicolored ray detecting part of his ear. 2) Original photograph fed into the algorithm.
Posted: 12 Nov 2010 10:51 AM PST
Using the Hubble Space Telescope and a cosmic magnifying glass effect, astronomers have put together one of the most detailed maps yet of dark matter in a giant galaxy cluster.
Dark matter is the stubborn, invisible stuff that makes up nearly a quarter of the mass and energy of the universe, but refuses to interact with ordinary matter except through gravity. The only way to know dark matter is there at all is by observing how its mass warps and tugs at visible matter.
When a lot of dark matter clumps together, as in massive galaxy clusters that contain hundreds or thousands of galaxies, it can act as an enormous magnifying glass for even more distant galaxies. The cluster's gravity stretches and distorts the light from galaxies behind it like a fun house mirror. Astronomers on Earth see multiple warped images of each galaxy, a phenomenon called gravitational lensing.
Gravitational lensing can give a good idea of how much dark matter is in a cluster, but up until now astronomers had to guess at where exactly the dark matter was.
Now, using an image from Hubble's Advanced Camera for Surveys, astronomers have built a high-resolution map of exactly where the dark stuff lurks in a galaxy cluster called Abell 1689.
"Other methods are based on making a series of guesses as to what the mass map is, and then astronomers find the one that best fits the data," said astronomer Dan Coe of NASA's Jet Propulsion Laboratory in a press release. "Using our method, we can obtain, directly from the data, a mass map that gives a perfect fit."
Abell 1689 lies 2.2 billion light-years away and contains about 1,000 galaxies and trillions of stars. By combining the Hubble image with earlier observations, astronomers picked out 135 multiple images of 42 background galaxies.
"The lensed images are like a big puzzle," Coe said. "Here we have figured out, for the first time, a way to arrange the mass of Abell 1689 such that it lenses all of these background galaxies to their observed positions."
Coe and colleagues superimposed the locations of dark matter in the cluster (shown in blue, above) onto the Hubble image. The results, which appear in the Nov. 10 Astrophysical Journal, confirmed that Abell 1689 has more dark matter packed closer together than astronomers expected for a cluster its size.
That extra bulk could indicate that galaxy clusters formed earlier in the history of the universe than astronomers thought. Dark matter's gravity pulls matter together, but it's countered by another, even more mysterious force called dark energy, which pushes matter apart. Once dark energy became an important player in the early universe, galaxy clusters would have had a hard time sticking together.
"Galaxy clusters, therefore, would had to have started forming billions of years earlier in order to build up to the numbers we see today," Coe said. "At earlier times, the universe was smaller and more densely packed with dark matter. Abell 1689 appears to have been well fed at birth by the dense matter surrounding it in the early universe. The cluster has carried this bulk with it through its adult life to appear as we observe it today."
More data is still to come from a project called CLASH (Cluster Lensing And Supernova survey with Hubble), which will aim Hubble at 25 galaxy clusters for a total of one month over the next three years.
Image: NASA, ESA, D. Coe (NASA Jet Propulsion Laboratory/California Institute of Technology, and Space Telescope Science Institute), N. Benitez (Institute of Astrophysics of Andalusia, Spain), T. Broadhurst (University of the Basque Country, Spain), and H. Ford (Johns Hopkins University)
|You are subscribed to email updates from Johnus Morphopalus's Facebook notes |
To stop receiving these emails, you may unsubscribe now.
|Email delivery powered by Google|
|Google Inc., 20 West Kinzie, Chicago IL USA 60610|