Face Recognition

Here is a news item on CNN about face recognition using a 3-D scanner instead of a regular 2-D photo.

For a fleeting moment, Mohamed Atta appeared on an airport security camera minutes before he boarded one of the planes which crashed into the World Trade Center on September 11, 2001.

Was there any way the camera or its operator would have been able to identify Atta as a suspect before he hijacked and flew the first of two planes into the twin towers?

Israelis Michael and Alex Bronstein think they have the answer.

The computer whiz-kids — 22-year-old identical twins almost impossible to tell apart — have applied a new technology to recognizing faces in a way that may yet revolutionize international security.

“I said it to them as a joke: if you succeed in building a system that can distinguish between the two of you, you’ll get (a grade of) 100,” said the twins’ professor, Ron Kimmel of the Technion Institute in Haifa.

“They succeeded and got 100. They are brilliant.”

The technology scans and maps the human face as a three-dimensional surface, providing a far more accurate reference for identifying a person than current systems, most of which rely on two-dimensional images, Kimmel said.

The product can potentially meet a wide range of security needs in a world shaken by the September 11 attacks and a series of bombings blamed on Osama Bin Laden’s al Qaeda network, of which Atta was a suspected member.

Kimmel and one of his former pupils, Assi Elad, had already developed the algorithms used as building-blocks for the face-recognition system. The Bronstein twins constructed a 3-D scanner, together with engineer Eyal Gordon, and applied the ideas to face recognition.

The twins and Kimmel say they want to turn the technology — registered for a patent in the United States — into a commercial product, with applications ranging from airports and border crossings to security zones and teller machines.

“We have a prototype and we saw the idea works,” Michael Bronstein said. “There is a hope that this will become a commercial product and allow all of us to feel more secure.”

Unique facial signature
The technology records the surface of a person’s face by scanning it with a series of light patterns and stores the data as a three-dimensional image in a computer.

Employing mathematical algorithms similar to those used in Internet searches, the computer measures the distances between a number of sample points on the facial surface.

The distances are then reconfigured as straight lines in a three-dimensional space, creating a new and abstracted image, or signature, of a human face built on precise mathematical calculations.

Kimmel and the Bronsteins say that this signature is more or less unique to a particular person.

The advantage of the system is its ability to compare facial structures as they appear in different poses or light conditions, variables which could distort a face seen as a two-dimensional image.

“One of my students calls it sculpting in numbers,” said Kimmel. “This kind of mapping makes it all invariant, or it is not influenced by our expressions. If we smile a little bit or we change our face a little, it will still be mapped into the same signature, the same kind of surface.”

The system could be employed at airports or border crossings where a 3-D security camera could scan passengers’ faces and compare them with a database of three-dimensional pictures of suspected criminals or terrorists, the twins said.

The technology would not work with existing two-dimensional images of suspects.

Facial signatures could also be embedded in credit cards or entry permits. People withdrawing money from an automated teller machine or seeking access to a secure compound could have their identity verified by an on-site camera.

A facial signature would be effective as a means of identification for more or less for the same number of years as a passport photograph.

Ageing, cosmetic surgery, significant changes to facial surfaces such as growing or removing a beard could disrupt the matching process.

Twin phenomenon
The Bronstein twins know first hand the importance of face recognition. During a recent interview, even Kimmel confused the young men, who share the same lanky build, wispy brown hair and steel-rimmed glasses.

The twins immigrated to Israel from Russia in 1991 and say they have always shared the same interests. As adults, Alex and Michael frequently complete each other’s sentences and begin statements with “we” even when speaking on their own.

“We always studied the same things and it was always connected to science, if sometimes indirectly,” said Alex.

Michael added: “We started working on explosives at a very young age, when we were about eight years old, and it lasted until we were about 13 when we built a bomb that was so powerful, we were scared of it ourselves.

“So we swore that we would stop that and as a reward, Mom and Dad bought us an aquarium with fish.”

The face-recognition project was assigned in a computer science course the twins took with Kimmel at the Technion, where they are studying for a masters’ degree in electrical engineering.

Asked if there was any way to distinguish between them, except for Michael’s shorter haircut, Alex said: “I must say that I’ve got a girlfriend.” Michael chimed in: “We don’t share those.”

This is close to my research area. But I am frustrated at the way the popular news media presents scientific news. For one thing, is this really the first time anyone has used a 3-D model for face recognition? I don’t think so. Secondly, the article does not really tell what the achievement of these guys was. The reporter just thinks it’s cool that identical twins are working on recognizing their faces. And where’s the information telling us how they differentiate between identical twins in their algorithm? What is the heading of the article about? What puzzle?

Here is a web page on the Technion website describing their work.

Range image acquisition is a crucial part of the 3D FACE technology. First experiments were performed using a laser range camera proposed and built by Zigelman and Kimmel. Acquisition time was about 20 second, with postprocessing taking several minutes.

For presentation on the Ministry of Science exposition (23 – 26 September, Jerusalem), a coded-light range camera was built. The new setup features very fast data acquisition (less than a second).

This means using a 3-D scanner to extract all the information about the 3-D shape of the face. This severely limits its application in a real crowded public place like an airport.

The main work done by these guys seems to be to convert the data from the 3-D facial scan to a form which is not affected by facial expressions and to compare the data for two different faces.

By Zack

Dad, gadget guy, bookworm, political animal, global nomad, cyclist, hiker, tennis player, photographer

13 comments

  1. I think that you could scan faces during (say) the security check, rather than when they’re in a crowd.

    Seems to me that the breakthrough here is the long-distance/rapid acquisition of a 3D model. In other words, a hardware breakthrough. The rest of it – pattern recognition based on eigenfaces and so on – is not so novel, as you probably know…

  2. Even there I am not sure. There has been work done using 3-D geometric models of the face using two or three cameras. If the acquisition of the image is in a controlled seeting, those techniques will work without requiring a range scanner.

    I am not saying it is bad research; it has its uses. However, I don’t see the practical benefits. Just like my research!

  3. Dr. G. Gordon did the similar work in 1991. So it is not novel. Besides, the recognition accuracy is not goog as they described.

  4. ravindra: If you google, I am sure you can find some research code for face recognition. I don’t have any publicly available code myself.

  5. “They succeeded and got 100. They are brilliant.” – Gulp…. Guess the recognition technology won’t be exposed to the general public for another few years?

Comments are closed.