Advancing the state of the art in 3D human facial recognition

Automated human face recognition is an important computer vision problem with numerous applications including security, surveillance, database retrieval, and human computer interaction. Over two decades of research in the area has resulted in a number of successful techniques for recognition of color/intensity two dimensional (2D) frontal facial images. However, the performance of these algorithms degrades severely when variations in facial pose, ambient illumination, and facial expression are present. Hence, achieving robust and accurate automatic face recognition remains a nontrivial open problem. Recently, researchers have proposed using 3D facial models for recognition to resolve some of these issues. Three dimensional facial models provide explicit information about the shape of the face, can be easily corrected for pose by rigid rotation in 3D space, and are scale and illumination invariant. Although 2D facial recognition techniques based on local facial features have been very successful, such techniques for 3D face recognition have been poorly studied due to a lack of understanding of the discriminatory structural characteristics of human faces. Other successful 3D facial recognition techniques based on rigid facial surface matching suffer from a very high computational cost which renders them inappropriate for realtime operation. The existing 3D facial recognition techniques are also unable to handle changes in facial expressions. We attempted to resolve some these open problems in the area of 3D facial recognition. Discriminatory features for face recognition are numerical quantities that vary considerably between individuals, yet are constant for different instances of the same individual. However, none of the previous 3D face recognition algorithms based on lo(a) (b)