Bad Idea: Using Facial Recognition to Identify London Rioters

London Riots

A group of volunteer techies have started a motion to identify London rioters using facial recognition. A google group has arisen to collaborate on identifying the London rioters using facial recognition technologies. Experts in the field are joining the force to crowd source on the cause. The intent of the group so far is to take the pictures of the rioters from print, digital and online media and then process and match them with the resources where they can be recognized, such as social media sites like Facebook and twitter. While the notion at first seems very helpful, it is also disturbing, insecure in most terms, and very harmful in reality. I can attest to that fact, being an image processing major in my master’s.

Though image processing is one of the hottest topics among researchers, and is being practically used to identify people, it is not without error. You might have seen facial recognition systems deployed at factories and offices and in criminal justice offices identifying people, but the scenario is totally different in this case.

Image recognition requires very clearly photographed pictures or videos with proper lighting and captured at different angles to train the recognition software, which then identifies a person with a certain error rate. London rioters on the other hand have hoodies hiding them, have their faces covered, and they are acting very violent. So the images arising on the media are blurry, incomplete, and hardly reliable. We also do not have more than one or two pictures of a person, with hardly enough evidence for a clear facial match. This makes it nearly impossible to identify individuals with any accuracy. Recognizing the rioters with the available pictures, and finding them on Facebook, twitter or government records would result in serious trouble for an ordinary resident of UK. The results will inevitably end up fingering the wrong people, exacerbating the already troubled residents of the affected cities.

Additionally, the notion of finding and recognizing people online points to serious risks for more and more people who have established an online presence.  As per my experience, most of the people in Europe and North America hardly have any security enabled on their online profiles with most of the pictures and data available to everybody. For the rest of us with tight security enabled, our profile pictures (which mostly are our real pictures) are available to anybody in the world. Either friend or enemy, finding you is just a few clicks away.

via thegadgets.net