| With the rapid development of science and technology, people are higher demanding for communications in the last decades. The video communications which based 3D facial will be a direction of communication in the future, and the key technique is 3D-modeling. The face modeling has made lots of progresses, but it has drawbacks that lower realistic, more time consumed and huge data, etc.In this paper, the existed methods of 3D facial modeling had be studyed, and focused on the methods that based on two orthogonal photos. After analysised the main steps and relational techniques, the following is researched. First of all, automatic extraction the key points. Due to the larger number of key points, and accord with a fixed order for extracting the key points, manual method is tedious and prone to error, so extraction method for the key points of automatic positioning was studyed. Under the modeling requirements, certain number of key points determined on the front and side photos respectively. According the order of the head, feature position and key points, all the requirements key points was detectioned.Secondly, the deformation method for the shape modeling. Three general facial models was introduced. And then the common interpolation methods was analyzed about their advantages, disadvantages and applications. Using the vertices on the Candide-3 face model and the coordinates of the key points which has been extracted from photos, sparse model was constructed. Combined with interpolation techniques, a personalized head model will be obtained.Finally, texture synthesis and model rendering. In the wavelet domain, the front and side photos was stitched. For the smooth and low-fidelity texture map, it need to be modified. And studyed illumination on different photos to reduce the illumination effects. Established the one to one coordinate mapping between texture and shape of the model, it may accomplish the model in vision, all the 3D face modeling work will be over. |
PDF Full Text Request