Characteristic contour-based three-dimensional face model recognition technique

Research on face recognition started in the 1970s. Many sophisticated techniques have been developed in the last three decades, and the performance of existing face recognition systems has been improved constantly. However, there are still two very challenging problems: changes in lighting conditions and head poses. Several approaches have been proposed in recent years. They are mostly based on 3D models with high computational costs. In this thesis, we investigate the feasibility of using only a small set of characteristic contours extracted from 3D face models. Specifically, we intersect a 3D face model represented in a polygon mesh with a set of cutting planes placed at selected positions. The intersecting curves are called characteristic contours. The 3D surface-matching problem is therefore reduced to a contour-matching problem. Furthermore, a scan lined z-distance algorithm converts a contour into a list of z-distance values. Thus, the contour-matching problem is further simplified to a simple one-dimensional z-distance comparison problem. Experiments using this new method on a small database containing 20 different face models are presented. Two of these face models come from identical twin brothers and we are able to use this new method to show that these two models come from different people. Our preliminary results show a very encouraging performance in both comparison accuracy and computational speed.