3D Face Reconstruction Based Face Recognition

3D Face Recognition technology draws more and more attention in recent years, which is one of the hottest topics in Computer Vision and Pattern Recognition field. Compared with traditional 2D face recognition method there are many advantages of 3D Face Recognition such as robustness of pose and lighting variation. But the 3D face sample is hard to gain which makes its widely application impossible. In this paper, we propose a novel 3D face reconstruction based recognition method trying to solve the pose variation problem by reconstructing 3D face model from single photo and using 3D model for recognition. The main work is as follows:Firstly, 3D face database standardization. Standardizing the 3D face database by pose and position correcting and data normalization with automatic correspondence method.Secondly, this thesis proposes a new face reconstruction method called Edge Map matching. The Edge Map is a set of 3D reference points adopting improved Genetic Algorithm for matching between reference morphable model and input images. This method can effectively reconstruct faces under any pose with high speed.Thirdly, this thesis introduces a 3D face recognition method based on iso-geodesic distance. The iso-geodesic distance of close points on 3D face models barely change when the 3D model have small deformation. Based on this theory, represent face model by iso-geodesic neighborhood sampling characteristics. This method is robust to facial expression variation and is computational efficient.Lastly, verify the proposed 3D reconstruction based recognition method. Research on 3D+2D face recognition method. Experiments are conducted to test recognition rate and pose variation robustness.