Font Size: a A A

Research On Face Pose Estimation Based On Single Still Image

Posted on:2015-06-06Degree:MasterType:Thesis
Country:ChinaCandidate:Z DingFull Text:PDF
GTID:2308330485490721Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
Face recognition is a key topic in computer vision. It has lots of important applications. Research indicates that environmental illumination and face pose have a great influence on the performance of face recognition. By using active near-infrared light to replace the visible light, we can solve the former problem. The latter can be improved by using face pose estimation techniques. The face pose estimation which based on a still image has important value in face recognition.Estimating facial direction in the world coordinate system from a still image or a video sequence is known as face pose estimation. Existing pose estimation methods can be divided into eight categories, such as methods based on geometric model, detector array model and so on. By analyzing of their advantages and disadvantages, geometry algorithm is proved to be simple, fast and real-time. This paper focuses on the geometric model pose estimation methods.A face structural model is built and its facial parameters are determined by the statistics of 42 people. According to the principle of rotating in three-dimensional space, we deduce the corresponding relationship of face model between world coordinate and image coordination. Two face estimation methods are proposed in this paper. One method is based on three points projection the other one is based on triangular distance. Using iterative method, we get a complete fast algorithm of face pose estimation to each method. Both of them only need three facial feature points, which is the least in existing geometric model methods.Experiments are carried out to validate and assess two methods. We use simulated data to verify the correctness and the robustness of the two methods. Finally, we compare our methods with other three classical geometric model methods on three face image databases. The results show that the two new methods this paper proposed are more accurate than the other three face pose estimation methods.
Keywords/Search Tags:Face pose estimation, face statistical model, 3D rotation, perspective transformation
PDF Full Text Request
Related items