Your Face Reveals More Than You Think

Distinguishing lies from truths has never been easy. A developing lie detection technology from the University of Bradford in the U.K., thanks to the R&D efforts led by Hassan Ugail, Professor of Visual Computing, leverages facial-recognition cues and is about to undergo a real-life trial.

A team led by Hassan Ugail, Professor of Visual Computing at the University of Bradford, combines an HD camera, a thermal camera and an algorithm to monitor changes in facial expressions, facial-temperature profiles and blood flow. After monitoring these traits in a group of volunteers, Ugail determined that certain expressions and changes were often associated with lying.

The standard HD camera captures subjects' faces, and the algorithm identifies facial landmarks. Facial expressions of subjects are classified according to a system, facial action coding system, developed by psychologist Paul Ekman. The camera and algorithm help monitor changes in expression frame by frame.

Similarly, the high-resolution thermal camera is used to detect the temperature profiles and blood flow patterns of subjects' faces, which the algorithm also helps monitor. Taken together with the facial-expression data, the algorithm eventually makes a decision about whether a statement is true.

This new form of lie detection deviates from the reigning polygraph lie detector because it does not require physical contact with a subject. Traditional polygraphs monitor blood pressure, pulse, respiration and skin conductivity — attributes that are difficult to monitor remotely. By using facial characteristics, it is possible to detect lies without contact, and perhaps, without a person's knowledge.

In fact, Ugail thinks that a hidden lie detector may give more reliable data. “We believe if this was done covertly, the results would be better. In all experiments, the facial temperatures of subjects always go up in interrogation simply because they are entering into a situation where they are being interrogated.” This issue would be minimized if subjects did not know they were being monitored.

When working with humans, Ugail recognizes the importance of dealing with variability in a population, and thus tries to establish a baseline for each subject. “For this test to work, ideally what we need is a baseline because every individual is different; we understand that. We need to calibrate our system to the individual. This allows us to identify the normal expressions of the subject and the normal blood flow pattern and temperature profile,” Ugail said. When forming the baseline, subjects are asked questions where they have no need to lie, such as a person's name. These questions help researchers establish a “normal” profile of the person in the absence of lying, against which subsequent statements can be measured.

This system currently has an accuracy level of 70 percent,but Ugail is aiming to increase the level to 90 percent. Given the variability in the human population, Ugail does not think that it will ever be 100-percent accurate. The new lie detector will be used as a tool to help humans decide whether they trust a statement.

No Silver Bullet
There are still a few kinks that need to be resolved. It currently takes a few hours for the results to come through, though Ugail sees that number coming down in the future. Additionally, this technology is focused, at the moment, on the proof of principle, that you can detect lies, rather than dealing with the issues that would arise from using this technology in the real world.

Ugail is optimistic about its basic lie-detecting functions, saying that it could be ready for interrogation purposes in about a year's time. Given the noninvasive nature of this technology, it has potential in arenas beyond traditional interrogation. At airports, for example, critical security questions are often asked as passengers pass through customs and immigrations — locations where authorities may want to know if passengers are lying. In fact, there are plans to pilot this technology at an airport in the near future. However, use in a more complex airport setting will likely take much more product development; equipment often works differently in a laboratory compared to an uncontrolled setting.

From the lab to Real Life
An anonymous airport security professional and Stewart Heffernan, CEO of OmniPerception, were asked to weigh in on bringing this biometric lie detection technology to market. They identified three important issues to consider.

Testing
With biometric technologies, testing and piloting are very important. Good laboratory preparation is often hindered by unexpected environmental factors. “We've found that a combination of testing the perimeters of the software or hardware sets the baseline , and testing with a real-world environment gives you real data,” said the anonymous source.

“When you walk into an airport or public arena, there are always environmental factors that you have not thought of when you deploy these things,” Heffernan agreed. “There are almost always things that crop up once you actually deploy the equipment in a trial situation or live environment.”

Marketing Hype
People have unrealistic expectations of what biometrics can do, because of movies and popular TV shows like CSI. It is important to not only create a good product, but to make sure customers understand the limitations of your products.

“People nowadays watch CSI and assume the impossible is quite easy. It's important, throughout the process when you're launching products and designing products, to realize what problem you're targeting, what problem you're trying to solve and making the product good enough for that particular market or need,” Heffernan said. “You need to make sure your customer understands what problem you're solving and that you can't solve all the problems.”

End Users
Airports have practical considerations about the usability of a technology, integration into its existing infrastructure and the final ROI. For example, does the algorithm limit the number of cameras you can have on a server, which then would require additional servers and increase costs? “ROI can be measured in many ways: reduction of staff, reduction of cost, improved detection or improved transfer of information. Just because I reduce or remove certain issues does not mean the cost of developing and installing such software or hardware will be lower than having the existing manpower to do the same task,” said the source that wished to remain anonymous.