Research On The Application Of Face Recognition Technology Based On 3-D Facial Reconstruction

Person identification has been one basic activity of daily work in modern life. Being a novel technology, biometric is playing an important role because of its popularity, stability and manipulability. It can satisfy the demands of national security, public safety and human machine interaction. Biometric has become a potential trend in identification area.Biometric includes face, finger print, iris, palm line, auricle shape and DNA etc which used in person identification. Especially, face contain the most important information of person identification. These information can be extracted more easily comparing to other information. Techniques of Face recognition based on two dimension image have a long research history. But one 2D image can not offer enough information which is used to person identification, so there is a distance between experimental performance and real life. We invent techniques of three dimensions to enhance the identification performance. 3D faces contain more feature information, such as shape. Furthermore, 3D face recognition has an advantage of pose and illumination invariant. With the development of 3D techniques, 3D face recognition will be the trend of face recognition.This thesis provided a detail analysis on the 3D Morphable Model which successfully resolved the problem of pose and illumination invariant. We applied cluster analysis techniques to 3D face modeling overcoming the disadvantage in current face recognition system based on 3DMM. Then we proposed a framework for cluster modeling of 3D face. 3D face similarity should come first in the framework. We reasonably defined similarity by Conic Affinity and Gradient Affinity. Experimental results revealed the feasibility and efficiency of our method. We also designed a novel recognition strategy for our new framework of face recognition system based on cluster modeling. Cracked the problem of similarity calculation between 2D image and morphable model, and designed pointed experiment to demonstrate the efficiency of our method. Finally we summarized our research works and provided the future work directions.