||Three-dimensional face modelling and analysis
||Skoglund, Karl (Informatics and Mathematical Modelling, Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark)
||Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark
||This thesis describes methods for modelling the shape, texture and appearance of human faces in three dimensions. These models provide a foundation for applications such as image segmentation and object recognition.
The models use a data set of 24 faces acquired using a 3-D scanner. To make the data useful in a statistical context, the representation of the objects must be unified, a process called registration. The registration method described here uses a set of nine manually placed points of correspondence on each face. These are used to form a dense correspondence of thousands of points. This makes sure the model is able to capture the subtle details of a human face.
The thesis presents the mathematical methods, with in-depth descriptions of key concepts, as well as the outlines of the implementation. The capabilities of the models are demonstrated in applications such as 2-D face segmentation and fully automated face registration.
The results are promising, but calls for higher quality and quantity of face data. The models were successfully used to automatically register new face scans, however the image segmentation, implemented using a simple optimization algorithm, fails in most cases. A number of suggestions for future work are given, for example the implementation of a 3-D active appearance model for face recognition.
||Supervised by Assoc. Prof. Rasmus Larsen
||Statistical Image Analysis; Shape Analysis; Shape Modelling; Appearance Modelling; 3-D Registration; Face Model; 3-D Modelling; Face Recognition; VTK
Creation date: 2006-06-22
Update date: 2007-09-11