A Computational Model of 3D Face Recognition |
||
|
||
|
DESCRIPTION/ABSTRACT: We evaluate the existing 3D face recognition techniques and propose novel, biologically motivated methods. A major weakness of the existing 3D face recognition research is the lack of fast and accurate registration, which is traditionally guided by a few anchor points (e.g. the eyes, nose tip, mouth corners). We first propose a novel statistical method (IMoFA) for flexible and automatic training of Gaussian mixture models, and then we use it in a novel feature-based algorithm for facial anchor point localization (i.e. landmarking). With this algorithm, we automatically locate the eye corners, the mouth corners and the nose tip on the face. We also propose a novel structural correction algorithm (GOLLUM) to evaluate the quality of landmarks and to help localization under adverse conditions (e.g. to detect the eye corners when the subject wears a sunglass). We compare GOLLUM with a recent competing method (BILBO), and show that it is more accurate. We test the success of automatic landmarking under rigid and non-rigid registration methods. For the rigid registration approach, we implement the popular iterative closest point algorithm (ICP), which iteratively aligns a test scan with a gallery face. The most important drawback of ICP is the computational cost of registering a test scan to each scan in the gallery. By using an average face model (AFM) in rigid registration, we show that the computation bottleneck can be eliminated. We propose a novel method of generating the AFM. We propose a shape-based clustering approach that assigns faces into groups with nondescript gender and race, thereby avoiding possible ethical concerns: The proposed system does not perform gender or race classification at any point. Finally, we propose a regular re-sampling step that increases the speed and the accuracy of the system significantly. Our results on the state-of-the-art FRGC database are very promising, and confirm our claim that registration is a critical issue for 3D face recognition. SPEAKER BIO: Albert Ali Salah is a research and teaching assistant in the Perceptual Intelligence Laboratory at Bogazici University in Turkey. He defended his Ph.D. dissertation in January 2007, and presented a paper on "Hidden Markov Model-based face recognition using selective attention" at the SPIE Conference on Human Vision and Electronic Imaging," in San Jose in early February. His research interests include:
|
