In this thesis I describe a new method to detect landmarks in 3D facial scans using a facial template model. Obtaining landmarks is a first step towards a face data processing at a semantic level. The use of facial landmarks has been used with relevant results in many application contexts like face animation, face recognition or face comparison. An advantage of using 3D scans as opposed to photographs is that more information is present in 3D scans in the form of a surface description. Moreover, 3D scans are also not affected by lighting conditions. Techniques used to find landmarks on photographs affected by one type of lighting condition, might not work on a picture with a different lighting condition.

A first step in the proposed landmark detection process is to register the facial template model onto the scanned data. When we fit both the shape and the expression of the facial model, we are able to achieve a more accurate registration of the facial model to the face scan. When we assign landmarks to this facial model and obtain an accurate registration of the model to the face scan, we can obtain the landmarks in the face scan in a rather trivial manner. The well known ICP registration algorithm has been adapted so that it is capable of the specific task of facial model registration.

A dataset has been build by scanning the faces of 18 subjects in 8 different expressions. Also a set of landmarks have been chosen. Results have been acquired by implementing the proposed methods and applying the landmark detection process on the facial scan dataset. For all facial scans, a registration can be found and the locations of landmarks can be found. A quantitative evaluation has been made by manually selecting the landmarks on the scan and calculating the difference with the locations that are found automatically. The results indicate that by changing the shape and expression of the facial model to acquire a better fit, a more accurate location is found for many landmarks and expressions.