Lecturer: Hello class, so today we’re going to be looking at facial recognition, and to the different sorts of technology that go into facial recognition.  Now before we start, can any of you tell me where we can see facial recognition in action? yes, you at the back?

Student: In the TV show Las Vegas?

Lecturer: Yes, well you’re right.  In this popular TV show, the security team pull images of the individuals from their surveillance system and run the image through a data base to identify the person. In that way, all the card counters and blacklisted gamblers can be escorted from the poker tables.  It looks easy on TV, but in the real world, facial recognition is a tricky business.  So let’s start with the more traditional methods of facial recognition.   Every face has peaks and valleys, and these can be translated into what is termed as nodal points.  Each face has about 80 of these, and they include distance between the eyes, the length of the jaw, the width of the nose, things like that.  These measurements can be used to create a numerical code, and this is called a faceprint.  This system is good, because it can compare two dimensional images, such as photographs.  The problem is that the images have to be controlled.  The person has to be staring straight at the camera, there must be no variance in facial expression or lighting, because any variance in these parameters reduces the effectiveness of the system.  So they had to come up with another way.

So the new way of recognising faces is by using a 3D model.  It has better accuracy, allegedly.  3D imagery detects distinctive features in the face, such as the curves of the eyes, nose and chin – features which do not change over time.  These are measured at the sub-millimetre level.  Interestingly, a 3D image can be taken not only from a live scan but also from a 2D photograph.  And another good thing about the 3D system is that it can recognise a person from a range of angles, the person doesn’t have to be directly facing the camera, as in 2D technology.  Once again, the system gives each individual a unique code – a set of numbers that represents the face.

It’s easy to match a 3D image to another 3D image, if you already have a 3D image in your database.  It’s less easy to match 3D images to 2D images.  But what they can do is pull certain measurements from the 3D image, such as size of the eye and so forth, and use this to convert the 3D image into a 2D image, and this image can be more easily compared to the 2D images in the database.

But it’s not just the measurements which can be used to recognise faces.  There’s also a new development called Skin Biometrics.  This uses the uniqueness of skin texture to get its results.  The process takes a picture of a patch of skin, and the system will then identify any pores, lines, moles, blemishes and other features of skin texture.  This method can be used to identify identical twins, something that cannot be done with the 3D technology.  Its other advantages over 3D imagery are that it’s insensitive to changes in expression, blinking, smiling and so forth, and can compensate for changes in facial features – such as the growth of a beard, or wearing glasses.  It’s not perfect, though, as it is sensitive to lighting conditions and poor camera resolution, and if there is glare from the sun.

So, now we’ve covered the main types of facial recognition, we’ll move on to its uses.  Now, has anybody here ...