Catalogue of Artificial Intelligence Techniques
Three Dimensional Face Recognition
Aliases: 3D, Face, Recognition
Keywords: 3d, face, recognition
Author(s): Thomas Drapier
Face recognition using three dimensions emerged in the late 1980s. It was invented to try and resolve the problems of regular two dimensional face recognition. The main problem with two dimensional recognition is when face expression, lighting and cosmetics changes from the original model, the algorithms have difficulties matching them with faces. Three dimensional face recognition tries to resolve that problem by taking into account the geometry of the face.
Face expressions are still a problem even with three dimensional face recognition since they can significantly change the structure of the face. One problem when involved with three dimensional problems is acquisition. Indeed, only expensive range cameras are able to give three dimensional ready pictures. Lately, solutions where regular gridded two dimensional pictures are applied on a three dimensional model have proven that regular cameras can be used with a good efficiency.
A way to treat the three dimensional face recognition problem is to use the identity of a person as the intrinsic geometry of his facial surface while the facial expression as the extrinsic geometry. Isometrically embedding the model of a three dimensional face into a low dimensional Euclidean space represents the intrinsic geometry of the face. Multidimensional scaling (MDS) is used to do the embedding. This gives a canonical form of the face which could be used to identify a person with different expressions. This is a fast and accurate solution. Another way of doing a three dimensional face recognition is to use a GMDS (Generalized Multi Dimensional Scaling) solution which can work when parts of a face are missing. It is supposed to be even more accurate. It uses a metric distortion to measure the differences between models and identities.
It has been shown that three dimensional face recognition algorithms achieve much better than regular two dimensional, and that it gets as efficient as fingerprint recognition.
- Michael M. Bronstein
- http://www.cs.technion.ac.il/~mbron/publications/BroBroKimAVBPA03.pdf Michael M. Bronstein
- http://www.cs.technion.ac.il/~mbron/publications/BroBroKimECCV06.pdf Michael M. Bronstein
- A. M. Bronstein, M. M. Bronstein, R. Kimmel, European Conf. Computer Vision, European Conf. Computer Vision, 2006.
- A. M. Bronstein.
- A. M. Bronstein, M. M. Bronstein, R. Kimmel, Expression-invariant 3D face recognition, roc. Audio- and Video-based Biometric Person Authentication.