Research On 3D Face Modeling

Nowadays, 3D modeling techniques are widely used. In recent years, with the rapid development of high-precision laser scanners and high-performance graphics workstations, the 3D face modeling becomes an active topic in computer vision and computer graphics.There are many techniques existed to model a 3D face. The 3D face can be reconstructed from a series of images. With the help of a generic model, the face can even be reconstructed from one image.We all know that the more information we have, the better 3D face be reconstructed. However, getting so much information is expensive in most cases. More importantly, it is difficult to collect enough information. For example, this may occur in the application of video retrieval for a particular person. So, this thesis discusses the method of 3D face construction based on a single frontal face image.First, we establish the face database of the frontal and lateral face information. According to the key features of human faces described in anatomy, we define the points and distances which need to be marked and measured, preparing for the next step. Then, using BP neural networks, input is frontal data and lateral data is output, finding the relationship of the two. With the weights we can estimate the depth values of the feature points. The convergent speed of standard BP algorithm is too slow. In order to improve it, an improved rapid algorithm for BP network is represented, on the basis of LM optimization algorithm by using double polarity S compressed function as transfer function. Numerical experiments show that the new algorithm can save CPU time effectively and improve the accuracy. Third, we define the feather points of the generic face model based on the measure points defined in the face database, and use the interpolation based on radial basis function to obtain special person's 3D face mesh. At last, the face model is more realistic after texture mapping.