Biometrics has become a popular and growing area of research in computer science due to its high reliability, and unfortunately due to recent social and political developments. In this area,  Image Processing methods cannot only be applied to fingerprints and iris scans, but also to images of whole faces.

Face Recognition:

The task of identifying a person from an image of their face is referred to as face recognition. The following application scenarios may be thought of:

- surveillance: tracking people using pictures of their faces

- access control systems

The Application:

The latter application scenario is the basis of our project:

-   a picture I is taken of a person requesting access to a door. Certain constraints can be assumed, e.g. frontal view, or a reasonable distance to the camera.

-   a database provides biometric information B on all authorized persons

-   an Image Processing system derives geometric information G from I

-   a decision-making system uses G and B to grant or deny access

Our Approach:

One way is to base classification on intensity information from the image itself. The problem with this approach is that it can be tricked by disguises like contact lenses, haircut, glasses, or a beard.

In contrast to this, the underlying idea of our approach is to identify so-called key points in the face – like the pupils or the tip on the nose – and exploiting statistical information about them. We believe that this approach is much more robust, or can at least be used to enhance conventional approaches.

Data Source:

As an underlying data source, we use the BioID Face DB (http://www.humanscan.de/support/downloads/facedb.php), which provides

- a varying range of images of several users

- coordinates of 20 face key points in each image

This made thorough testing and rapid prototype development possible.