Background

One of the oldest forms of identification is photo ID.  What is it about the face that makes it easy to recognize?  Like most parts of the body, the face has detail that can be used for identification, but more importantly, the face is usually visible and not covered by clothes.  We have to see and breathe, eat and talk, so we usually don't cover our eyes or nose or mouth.  We can also see facial features from a distance.  Compared to other biometrics, face images are the most accessible (easiest to collect).

What makes a face hard to recognize?  Celebrities and criminals often use disguises to keep from being identified.  Hair is not a reliable feature, since the style or color can be changed, or hair can be covered with a hat.  Similarly, eyebrows can be shaped or colored with pencil, or covered with glasses.  In general, makeup can change appearance.  Also, appearance changes with age.  For these reasons, as a biometric, face recognition is the least reliable. 

While the accuracy of face recognition is not fool-proof, it is still a useful tool.  For instance, think of the number of security cameras.  The UK has so many security cameras that travelers may be visible on CCTV 300 times per day.  Not all of the views are useful, but the trick is to detect a face in a good position and analyze that image.  A computer program can detect a face by looking for a pair of eyes, for example.

What if we had a three-dimensional model of the face?  A special camera, called a range camera, can capture range measurements and construct a 3D face model.  Another advantage of a range camera is that it is not affected by changes in lighting.  This is a newer technology, that can improve performance for face recognition. 

What things about the face cannot be easily changed?  The location of the eyes, nose, mouth and chin depend on the skull and jaw.  If a face is level with eyes open and mouth closed, these locations can be easily compared.  Since bones are internal and rigid, these measurements can help identify a person.

Then, we use measurements and math to compare faces.  What are some measurements that would be useful?  Some typical measurements are:

• distance between the eyes
• width of the nose
• depth of the eye sockets
• shape of the cheek bones
• length of the jaw line

Face recognition is useful for crime solving, but it also has commercial applications.  Have any of you used Picasa for sharing pictures on the internet?  It has face detection, and if you label some pictures of a person with a name, it can find other pictures or that person.  Pretty cool, huh?  Some photo editors also include face detection to let people create wacky images for fun.