| 3D face shape biometrics, with greater pose and lighting condition data invariance than 2D (photometric), has the potential to yield superior performance to 2D data for some applications. However, many of the capture limitations of 3D scanners are the same as those of 2D biometric capture devices with respect to lighting, environmental, and subject configurations. Many of the claimed advantages of 3D over 2D do not exist under current capture configurations. In addition, 3D scanners themselves are more expensive than 2D cameras, and 3D biometric data are unavailable on many of the subjects we would like to identify. Further, the significant computational cost of 3D face recognition has made large scale deployment of 3D face recognition impractical. The focus of this thesis is to address these issues to improve the feasability of 3D face recognition so that it is more applicable outside of a research environment. In particular, the focus is on improving the methods and hardware needed to produce a 3D model of a face, improving biometric recognition and verification performance, and decreasing the computational cost to allow for larger scale applications. In this thesis, I propose a new structure from motion approach, a new fast 3D face biometric, and examine the impact of movement on existing structure from light devices. |
