Research Of 3D Face Recognition Based On Facial Landmarks

Face recognition technology is an important part of biometric recognition technology, and also has great value in many fields. The 3D face recognition compared to 2D face recognition has more broad application prospects, because of illumination and pose invariance. But 3D face data acquisition is more difficult than 2D image, how to select the useful information from the 3D data is a key for 3D face recognition. At the same time,3D face recognition is also affected by expression variations, how to overcome the influence of facial expressions is an important research direction of 3D face recognition. Effective solutions are proposed in order to solve the above problems, the main work is as follows:1) A coarse to fine algorithm of 3D landmarks location is proposed in order to achieve effective recognition features extraction. In the thesis, the ASM algorithm which is used for 2D image is applied to 3D facial data. First, the 3D facial point clouds are converted into the depth images through interpolated projection, then the ASM algorithm is used in depth image to realize the rough localization of 7 landmarks including the tip of nose, inner and outer corners of eyes and the corners of mouth. Next, the Shape Index algorithm is used for accurate localization of the 7 landmarks. Finally, Three experiments are conducted, and the results show that the algorithm of 3D landmarks location is precise and real-time.2) A 3D face recognition algorithm which is based on landmarks is proposed in order to solve the problem of facial expression variations. First, some necessary preprocessing works are conducted before the features extraction, and the landmark which is located in the middle of nose tip and nose root is defined. Next, a set of facial curves with different geodesic distances from the landmark(located in the middle of nose tip and nose root) are extracted to represent a facial surface, and then Procrustean features (distances and angles) defined on pose-invariant curves are elected as the final recognition features. Finally, classifier fusion in decision level and fusion in feature level are compared respectively, and the final result is given.3) A 3D face recognition system is organized which is based on the above research on the 3D face recognition technology, combined with the 3D measurement system developed in our laboratory.3D face recognition system is a set of simulation of the access control system. The system not only has the function module of recognition, but also has some basic functions such as user management and attendance.