Oxford Uni develops porky-pie detection software

Being able to read a person’s expression is a useful skill, whether you are in a high stakes poker game or predicting which team’s paying over the odds haggling on Bargain Hunt

Unfortunately, it’s often very difficult to successfully determine when someone is telling porkies, and demanding to rig people to an old fashioned lie detector might be OK on the Jeremy Kyle show but it’s unlikely to go down well elsewhere.

However, small and fleeting facial expressions can sometimes give the game away when a person is attempting to hide their true emotions. 

Researchers at Oxford University reckon that such ‘micro-expressions’ can be tracked by software developed in partnership with Oulu University.

According to Tomas Pfister, an Oxford University engineer involved in the study, micro-expressions are incredibly rapid, and last only between a twenty fifth and a third of a second. 

It is, in some cases, possible to train people to catch those expressions, with US airports officials instructed to look out for giveaway signs in passengers when questioned. Typical giveaways for TSA officers are spotting people with ‘brown skin’ or when a parent refuses to let them strip-search kids right there in the queue.

Those indicators are quite broad, though, so such lie detection is about as intuitive as spotting a poker bluff from a seasoned pro. Oxford University’s automatic detection should kick those problems out of the door, significantly reducing the guesswork involved. 

One of the main problems for detecting fibbing micro-expressions is that they quickly vanish from the face. With normal speed cameras, tell-tale expressions will only appear for a couple of frames that can be examined by a computer. The researchers are using a temporal interpolation method to fix this. It involves filling in gaps with existing data, meaning that even a standard camera could pick up on lies.

The automated method was able to pick up on micro-expressions with a “significantly” higher rate of success than typical face reading, with 79 percent accuracy possible at this point in development.

While the expression detection might not provide conclusive evidence of lying, the researchers believe their approach can distinguish between deceptive and truthful micro-expressions.