Multi-pose Face Recognition Based On 3d Face Model

Face recognition is one of the most popular research fields at present, and is widely applied in many areas, such as information security, video surveillance and video tracking. Recently, most researches concerning face recognition are based on 2-D face images, but because of the influence of illumination, pose and expression variation, the improvement of recognition rate of 2-D face recognition is greatly impeded. This makes it still difficult to build a robust face recognition system up to the present day. 3-D face model possesses more rich information than 2-D face images. Therefore, implementing face recognition on 3-D face model is one of the effective approaches to tackle the present problems. The key part of 3-D face recognition is building the corresponding model with the acquisition of the shape model and accurate texture mapping as the core issues. This problem is still not resolved ideally in face reconstruction area now.In this thesis, we study the image-based and model-based human face reconstruction methods, create a new method for reconstruction, and discuss about the applications of the model in several hot fields. Several specific research points are showed as follows.(1) RBFS and B spline method are used to 3-D face constructionStereoscopic sense of vision is used to reconstruct 3-D feature points. Then multi-layer and multi-region RBF (Radial basis function) based on feature points of frontal and profile face image is adopted for reconstructing the generic face model to get 3-D specific face model with visible organ characters. Moreover, B spline is used in 3-D face curve reconstruction. Then the near face global texture is mapped into the reconstructed specific face model and the 3-D realistic specific human face model is obtained.(2) Face pose parameter is estimated by calculating the three freedys of facial rotationOn the basis of traditional geometry model, according to the theory of image graphy geometry, a new method which doesn't need any other special device is presented, which uses the features of personal facial configuration and only one image to estimate face pose parameter. Face pose parameter can be estimated rapidly when three freedys of facial rotation are calculated separately.(3) 3-D face model is used to multi-profile face recognitionAt first, RBFS and B curve method is combined to carry on 3-D face reconstruction; then face pose parameter can be estimated rapidly when three freedys of facial rotation are calculated separately; later, the face image's position parameter and the 3-D face model is combined to produce new face images which have the same position parameters with the matching one; at last, 2-D face recognition is done with the same face positions.