Face Verification (2008-2010)
A video of your face contains useful information such as who you are, where you are looking and how you are feeling. If we can extract this information from the video, it potentially paves the way for automatic face verification (i.e. determining whether you are who you claim to be), augmented reality and robots that can recognize emotions.
To get this information from the video we track landmarks on the face (e.g. around the eyes, nose, mouth and jawline) using a technique, known as the Active Appearance Model, that was pioneered by Tim Cootes and Chris Taylor at the University of Manchester. The positions of these landmarks relative to each other define the shape of your face and indicate where you are looking, your expression and (to a small degree) who you are. Your identity, however, is more strongly indicated by the texture of the face which we can extract by cutting out just the bits of the image that lie within the boundary defined by the landmarks on your eyebrows and jawline.
To register with a face verification system, you first record a video of your face. To log in to the system, you then record a new video and the system compares the new video with the original one to see if they match. If, however, you are facing in a slightly different direction or have a different expression in the two videos, your appearance on the screen is very different which may confuse the verification system.
To avoid this problem, we use the tracked landmarks to rotate, scale and stretch the image of your face so that it seems more like you are looking straight at the camera and have a neutral expression. This way, most of the remaining differences are down to who you are which ensures that the verification process is more accurate and secure. Ideally, using a biometric such as your face will eventually replace passwords and PINs that are difficult to remember and are easily forgotten or stolen.
Mobile Biometrics (MoBio)
In the Mobile Biometrics (MoBio) project, we aim to do both face and voice verification using the camera and microphone that are built into a modern mobile phone (see the video above). This will provide secure access to mobile internet applications (such as e-mail, social networking and internet banking) without the need to remember passwords. Getting the computer to 'remember' (i.e. store) your passwords is especially dangerous since your phone could easily fall into the wrong hands.
The MoBio project is unique in that it combines face and voice verification algorithms that run on the mobile phone (rather than a high-powered PC) and must work with the low quality data that we get from a typical mobile camera.
Frequently Asked Questions
Won't the system be fooled by a photograph?
No. Because we track individual features on the face, we can check that the face is 3D, that the lips move and the eyes blink and so on. Also, the system uses both face and voice recognition so you would need a recording of my voice also. Playing a video clip might be able to fool the system - that is the subject of a follow-on project that is about to start at our project partner, IDIAP.
Hasn't this all been done before?
To the best of my knowledge, this is the first example of detailed facial feature tracking on a mobile phone. There are plenty of examples of face detection on mobiles but our system gives a much greater level of detail in the tracking, such as being able to track rotation, even through a full 360 degrees.
Impact
This work has attracted widespread publicity across various channels (search for "Tresadern face"), including these selected examples:
| Daily Mail Online | The Times of India | The Hindu Online |
| Slashdot | Engadget | Tech Encounter |
| Wired | PC Pro | GizMag |
| TechEye | The Engineer | SciTech Report |
| Eureka! Science News | how-do | Security Technology News |
| IntoMobile | RCR Wireless | Mobile Computing News |
| the Nokia Blog | Nokia-N900 | Nokia N900 Applications |
| Nikkei Electronics (Japan) | Dantri News (Vietnam) | MaemoBrasil (Brazil) |
| L'Atelier (France) | DiePress (Germany) | Aplicaciones Nokia (Spain) |
| DailyMobile (Sweden) | MyMaemo (Czech Republic) | Dnevnik (Slovenia) |
| BBC Radio 5 Live | BBC Radio Manchester | BBC Radio Newcastle |
| Real Radio Manchester |