Facial Feature Scans

From SCAN THIS NEWS 11/11/98

Some people will argue -- after reading the following article -- that systems designed to electronically "recognize" a face in a crowd will never be reliable. They will contend: such systems will never be 100% accurate. Unfortunately, "accuracy" is a very subjective term, and, if the administrators of these systems,(i.e. the ones who can forcibly apprehend the "identified" subjects), deem their system accurate, then the burden of "proving innocence" suddenly becomes that of the accused. At that moment, it no longer matters whether the system was or was not "accurate" -- you're under investigation.

From the 11/1/98 Sunday London Times www.sunday-times.co.uk - you must register at the site and then search by date.

Scanner picks out criminals

TRIALS of a computer system that can recognise criminals and terrorists begin this week.

The equipment is to be hidden in an unnamed British airport to see if it can pick out criminals even if they have grown a beard or are wearing glasses to hide their identity.

The system is the first that can produce three-dimensional scans of a face instantly and then search a database of suspects for a match.

It has been devised by Cambridge Neurodynamics, which already produces software that analyses fingerprints and then searches a database for a match. Police in South Yorkshire use its software.

Cambridge Neurodynamics says its facial-recognition project will enable security officers at Britain's ports and airports to concentrate on people the computer indicates bear a good likeness to known criminals and terrorists. The alternative is to rely on trained officers to remember faces of people on the wanted list.

George Harpur, a Cambridge Neurodynamics systems consultant working on the project, says a computer will be able to store thousands of pictures of wanted people.

"It should be like having an extra security officer with the most incredible memory," he says. "We're not intending to replace immigration officers but rather give them a tool that can point them towards the people they are mostly likely to be interested in."

The system works by taking pictures of a person as he approaches a video camera. The resulting handful of frames give several two-dimensional pictures. To add depth to the face, two low-power lasers scan its contours from either side.

A computer combines the contour information with the images it has of the front of the face. It can then build a virtual model of each traveller's face, which is checked against the database of wanted people.

"The airport involved in the trial is particularly interested in the possibility of using the technology to track people who bring in groups of illegal immigrants," says Harpur. "The computer could be alerted to their presence and take a 3D scan of their faces when they pass through passport control. Then the system could track them every time they enter the country and build up evidence against them."

The system can also be used to store 3D Photofits of terrorists. Again, it is most likely to be used to take a 3D scan of suspects when they pass through its lasers for the first time. The scan would be stored and the computer would log the suspect's movements.

Cambridge Neurodynamics says it can build 3D Photofits from ordinary photographs or video stills but says the system is far more accurate if the suspects are photographed by the system.

The team believes 3D scans are necessary because 2D technology can be fooled. It generally measures the distance between facial features but these can change with the angle of the head to the camera.

The new computer system can be fooled only if a suspect has surgery to change the shape of his face. Criminals who realise they are about to be scanned and look away or pull a funny face will not trick the computer. It is geared towards seeking similarities around the eyes, which humans use to pick out one another and that do not change with expression and age as much as the rest of the face.

The airport trial will last six weeks. If the equipment works, the company will load the software on a powerful workstation computer, rather than a PC, so searches can be speeded up.

Cambridge Neurodynamics also plans to approach banks to set up a trial in which the computer would allow only designated staff into security-sensitive areas. Eventually, the technology could be used to verify a customer's identity before he is given money.

Return to ID Schemes - Click here

Tuesday, October 13, 1998 - UK

Candid camera for criminals: Three steps to criminal-spotting

A revolutionary surveillance system that can pin-point known criminals as they walk along in a crowd is being put to the test in a London borough.   The Mandrake face recognition system will seek out 'target faces' in the closed circuit television (CCTV) footage in the Newham area from Wednesday.

Councillor Ian Corbett said the decision to go ahead with the 60,000 six-month trial was made in response to the concerns of local residents.  "We've done surveys recently and 60% of the population said that crime prevention was the No 1 issue in the community," Mr Corbett said.

But fears of innocent people being identified by mistake have lead civil liberties groups to condemn the system and call for it to be tightly regulated.

Making a match

Newham has a network of 140 street cameras as well as 11 mobile camera units.   Images beamed into the council's security centre in East Ham will be compared with a database of target faces supplied by police.  The system can isolate the targets from the crowds of people appearing on CCTV. When a match is made the computer highlights the target and sounds an alarm.  An operator then checks the image and decides if it is necessary to contact the police.

The police in the area see the system as a way of making CCTV more efficient.   "The people who go onto the system will be convicted criminals," said Chief Superintendent David Armond.

Depending on the success of the trial, other targets like paedophiles could also be scanned. The system could also be used to help track down missing persons.

Advanced technology

Mandrake is the first identification system to be able to work from moving pictures. It has been designed by Software Systems International which has been concentrating on identification systems for several years.

Less advanced systems are already in operation, including one which compares pictures of criminals with individuals crossing the Mexican border.

One state in the US is using a database of millions of pictures to check on people who may be entering into more than one marriage, and another state is checking for duplicate drivers' licence applications in the same way.

"The ability to capture a moving face is quite a new innovation and makes a lot of difference to being able to work with things like CCTV," Software Systems marketing manager Pat Oldcorn said.

However she acknowledges that the computer will not always strike a perfect match.

"We do expect that we will get a little bit of difference in interpretation because sometimes it will pick up a face at a three-quarters angle. We will need to use the human element to check the authenticity of the picture," she said.

Big brother concerns

It is the risk of error that has the civil liberties group Liberty most concerned.

"The accuracy of facial mapping is very limited," campaigns manager Liz Parratt said.  "For example, you need only to look at a handful of photos of celebrities to see how different the same people can look in different photos."   "The claim that those who have nothing to hide have nothing to fear is rubbish. What the police call an 80% success rate is what we would call a one in five chance of a mistake."

Ms. Parratt said that even if the system did work, it would have to be carefully regulated to protect people's privacy.

But Councillor Corbett said he was most concerned about the civil liberties of innocent people.

A reduction in the crime rate in Newham over the next six months will persude the Labour-dominated council to continue with the system.

Return to ID Schemes - Click here