This web site is no longer maintained and the content may be outdated.
Please visit for up-to-date information.
No upcoming events...

Home / Graduate / M.S. Theses Completed
  Mustafa Okan İrfanoğlu, 2004    

Thesis Title

3D face recognition from shape information based on 3D surface registration


Face recognition techniques have shown a great improvement in the last decades. Although important advances have been realized on image based 2D face recognition algorithms, there still remains some challenges to deal with like illumination, pose and expression variations. A convenient way to deal with these problems would be to utilize 3D information where the shape information of faces is directly available.In this work, we propose two different methods to represent shape information of human faces to be used in a three dimensional face recognition system. The first one describes the faces with implicit polynomials and their invariants. The second approach, called the Point Set Distance method, is based on the registration of facial surfaces and establishment of point-to-point dense correspondence. The recognition performances of the algorithms are compared with other shape descriptors and our results show that the surface description method based on registration produces yields classification results as good as 98.9 per cent on noise--free database consisting of 30 people and 93 per cent on a noisy database of 106 people. The preprocessing and feature extraction steps are performed faster than the existing methods and the outputs of the registration process are ready to be used for statistical purposes such as Point Distribution Models or other shape variation analysis. A feature selection method is also implemented to investigate the importance of different facial regions in the human 3D face recognition process. Results have shown that the upper part of the faces plays a crucial role.
Boğaziçi University Department of Computer Engineering
Address: 34342 Bebek, Istanbul, TURKEY
Phone: +90 212 359 4523-24 Fax: +90 212 287 2461
general information:   webmaster: