杏吧原创

THE response was as aggressive and swift as the riots themselves. Within a few hours of the worst of last week鈥檚 looting across London and other English cities, attempts were being made to use CCTV footage to track down the individuals who had plundered shops and destroyed buildings.

But those raised on a diet of TV police dramas who expected crack law enforcement teams to simply plug the footage into a computer and then print out a list of suspects are going to be disappointed. The poor quality of most CCTV footage makes it almost impossible to trust standard automated facial recognition techniques.

One of the most common methods used to help identify an individual from camera footage is photoanthropometry, which uses 鈥減roportionality indices鈥� to compare a picture of a suspect on a police database, say, with a CCTV image. Key points on a person鈥檚 face 鈥� such as the chin, edge of the nose, or centre of the top lip 鈥� are marked and the distance between them measured. Someone experienced with this technique can then judge whether the two faces match.

Reuben Moreton from the London Metropolitan Police鈥檚 Digital and Electronic Forensic Service tested the techniques on poor quality, low-resolution footage of 13 volunteers. It produced 鈥渃haotic, inconsistent results鈥� (Forensic Science, ). The lack of facial detail reduced the precision of the PI measurements so that individuals with similar PIs could easily be confused with one another.

Imperfect CCTV and poor- quality images can easily thwart more advanced, automated techniques too, says Lyndon Smith at the Machine Vision Laboratory of the University of the West of England, Bristol, UK.

鈥淭here is an inherent reliability problem in conventional face recognition systems,鈥� he says. 鈥淐hanges in lighting, image quality, changing background and orientation 鈥� even make-up can fool them.鈥�

Smith鈥檚 group is working on a system called Photoface that he believes could help. The system takes a number of 2D images of someone鈥檚 face and then stitches them together into a 3D model in which the lighting can be manipulated and the face can be viewed from different angles. He envisages Photoface being used on CCTV footage fed to a police control room. 鈥淭his is a system that could help with the identification of people in unusual situations or low light,鈥� Smith says.

, senior technology consultant at IT security firm Sophos, agrees that current systems are not up to the task. 鈥淭his is not CSI Miami,鈥� he says. 鈥淐omputers can鈥檛 do the things you see on TV shows.鈥�

While a Metropolitan police source told New 杏吧原创 that face recognition technology is being used in some cases, it is only effective for face-on shots. It鈥檚 far more useful, he says, to post CCTV images of suspected rioters on the Met鈥檚 Flickr account, and then invite the public to comb through the images and point out suspects they recognise. 鈥淲e can get a JPEG in front of a million people in 2 minutes. No computer system can match that at the moment.鈥�

聯Face recognition works for full-face shots but a JPEG posted on Flickr can reach a million people in 2 minutes聰

Some digital vigilantes believe face recognition technology will provide the answer. A group of programmers have formed the Google Group 鈥淟ondon Riots Facial Recognition鈥�. They won鈥檛 discuss their techniques, but told New 杏吧原创: 鈥淲e鈥檙e just a bunch of computer programmers who want to use technology to try and help the situation.鈥�