Could Facial Recognition Help ID Jihadi John?

Facial recognition software could be used to help identify the Islamic State militant thought to have killed a Briton and two Americans in propaganda videos.

Footage has been released over the last month apparently showing the murders of US journalists James Foley and Steven Sotloff, and British aid worker David Haines.

The IS militant - who speaks with a southern English accent and has been dubbed Jihadi John - shows only his eyes in all three videos, none of which have very high resolution.

Hugh Carr-Archer, chief executive of Aurora Biometrics, a facial recognition firm based near Northampton, told Sky News: "To be frank, the video itself is not going to be very useful.

"Because once you start blowing it up, looking at fingers, looking at eyes, you lose definition.”

However, partial face recognition is an established technology.

The software is not dissimilar from full-face recognition and treats the suspect's face as a 2D image.

It then identifies the features, picking out points on the eyes, nose, eyebrows and chin, among others.

It analyses their relative position to create a point map, then matches that map against a database. It can match a million faces per second.

With partial face recognition, the visible part of the face is extracted - in this case, the rectangle of the balaclava.

That small section is mapped to a generic face, which can also be matched. The approximation is more hypothetical, and rougher, but can still be useful.

Canadian firm Face Forensics specialises in partial face recognition, and its software powers the UK's ChildBase system - used to identify paedophile suspects.

Chief executive Iain Drummond told Sky News that for the system to work by itself, the video would ideally need double the resolution. But it could still be used as the basis for more advanced analysis.

"At that level you’re just not getting enough information to differentiate one face from another," he said.

"There are tools of course that can enhance images so if you can enhance that image itself, then obviously the face recognition system will have a much better chance of coming up with a likely match."