Should Big Brother get better vision?

Posted by s.hettrick on 11 October 2011 - 11:44am

Riot.jpgHeather Packer, one of our Agents, asks whether current facial recognition software could help identify  criminals.

The recent London riots and the BBC's Crime Watch special, which focused on video footage of the rioters so the public could help identify them, led me to think about whether there was a feasible automated solution using facial recognition software.

There are many problems that prevent a face recognition algorithm from successfully identifying a person from a digital image. The occlusion of facial features such as the nose or eyes from images, different headwear such as hats and glasses, and varying lighting conditions all make recognition difficult.

Imran Naseem's paper "Linear Regression for Face Recognition" [1] addresses these problems by dividing an image into segments. These segments are then analysed to identify and evaluate facial features, such as the corners of the nose and eyes. The novel idea is a distance-based evidence fusion algorithm, which provides a measure for the distance between facial features. This distance evaluates segments used to identify a person's face and rejects segments from the recognition process if the distance between the features is too great. This increases the accuracy of the recognition and makes the algorithm more effective than the state-of-the-art. The algorithm is backed up by statistics from five standard databases for face recognition, which contain photos of people wearing scarves and photos taken under different lighting conditions.

In the real world, facial recognition isn't widely used because reliable identification generally requires many photos of a person, and better quality images than those taken by CCTV and mobile devices. However, there are an increasing number of personal photos online, and the quality of images and facial recognition algorithms is improving. Naseem's software makes face recognition more accessible because it handles partially obscured faces, and it is faster and more accurate than previous approaches.

Facial recognition software can be used as a crime fighting tool to identify suspects or witnesses. While the police already use social media, such as Twitter and Facebook, to manually identify suspects, they could go one step further and use the images from Facebook accounts, which are already tagged with peoples' names. This could lead to a database of images that could match video footage automatically. At the time of writing there are over 750 million users on Facebook, who on average publish over 30 billion posts including web links, notes and photo albums, each month. Photos could be searched according to whether the owner lives near to the incident, to reduce searching time. The London Riot is a rare case, but this software could be used to help after sporting events which contain large crowds. Of course, the question of how to carry out facial recognition is different to the question about whether we should carry it out. But that's a discussion better left to the political arena.

[1]: Imran Naseem, Roberto Togneri, Mohammed Bennamoun, "Linear Regression for Face Recognition," IEEE Transactions on Pattern Analysis and Machine Intelligence, pp. 2106-2112, November, 2010.

Share this page