The growing need for effective biometric identification is widely acknowledged. Identifying an individual from his or her face is one of the most non-intrusive modalities in biometrics. Major challenges to face recognition system robustness include illumination and pose variations. This work introduces foundational research addressing two-dimensional intensity, infrared, three-dimensional, and multi-modal face recognition.
I contribute to the growing body of work surrounding face recognition by examining novel approaches to face recognition beyond the traditional, two-dimensional intensity modality. My infrared research made strides in overcoming the illumination challenge because this modality proved robust in the face of varied lighting conditions. Although infrared is invariant to changing lighting, infrared is not robust in the face of varying pose, it produced very low-resolution images with unreliable face registration when compared to their 2D intensity counterparts, and, in many cases, it would be cost-prohibitive.
In looking to overcome these bottlenecks, 3D recognition was the next, logical step given that previous trials with the three-dimensional modality demonstrated that it provides greater accuracy than any of the two-dimensional modalities. However, available 3D state-of-the-art scanners such as the Minolta Vivid series prove cost-prohibitive, are fragile, require that a subject remain immobile for several seconds, and are generally too intrusive for many real-world acquisition scenes. These observations provided the motivation for a proposed three-dimensional recognition system that is built upon a two-dimensional framework. I extended my infrared research to consider a three-dimensional recognition system that had a two-dimensional foundation.
This dissertation explores the possibility of using cost effective, flexible, accurate, and user friendly multiple-view stereo photogrammetry to reconstruct the three-dimensional shape of the human face for improved recognition performance. Specifically, I developed a novel approach to face recognition that relies on 2D images to successfully reconstruct 3D shape of the human face. This approach ultimately outperforms 3D shape obtained from a commercial scanner. This is noteworthy given that our approach does not require strict calibration as in the case of the commercial 3D scanner. Also significant is the demonstrated flexibility of this system to successfully perform 3D recognition on a database acquired originally for 2D face recognition.