Biometric facial recognition uses a camera to capture a person's face and compare it against one or more previously saved facial images. The first step is the digitisation of the image, e.g. using a PC. The recognition software then detects the face and computes its characteristic features. The resulting template from this computation is compared with the templates of the saved facial images. This is only ever not the case if the original image is used as a reference image, which is compared against a current original image for the recognition process.

There are different approaches to facial recognition, although all of them use certain key elements. Most facial recognition procedures use a digitised image to determine the characteristic facial features. The ones mainly used are facial characteristics that don't constantly change due to expressions, i.e. upper edges of the eye sockets, the areas around the cheekbones and the side parts of the mouth.

In general, the process involves a comparison of the characteristic facial features with the corresponding reference characteristics using classic image processing and image analysis methods, such as – after detection of the eyes – computing the facial characteristics on the basis of a grid that is positioned over the face. The template size can be up to 1300 bytes. The Eigenface method is a special category of biometric facial recognition, which is mainly used in the field of personal identification. Lastly, the first (research) approaches for 3D facial recognition now exist.

A general overview of facial recognition is available here (PDF).