Here's some science news to put a smile on the faces of those nice folk over at the department of homeland security (or maybe not). In a brief note in the journal Science, Rob Jenkins and Mike Burton of Glasgow University managed to boost off-the-shelf facial recognition software performance from 54% to 100%. This particular recognition system is used by Australian customs, amongst others. So what sort of tinkering with the software algorithms did they have to undertake to manage this you may ask? Absolutely nothing. These clever chaps simply created "average" faces for each of the people they were trying to recognise. That is, they took all of the photos they had of a person (different angles, different periods in their lives, different lighting) and morphed them all together (slightly more complicated than this, but not much). They then presented this averaged face to the same recognition system and, lo and behold, accuracy reached ceiling. Such an improvement in facial recognition software is unparalleled and will have interesting ramifications if it is widely applied to systems currently in use around the world. On the one hand, no more false alarms or facial mismatches. But on the other hand, security services will have less call to "randomly" pull people aside because they look a little bit like someone who they think might just possibly be involved in dodgy dealings. Having this option removed is not something they will enjoy.
The paper can be accessed here (password may be required).