Research Of Feature Detection And 3d Face Reconstruction Based On Morphable Model

Creating high realistic 3d face model has been a challenging problem, also due to the widely used of 3d face model, more and more attention is attached to it. Because of the direct use stereo vision to set up 3d face model, the effect is not more accurate, and can be based on the error of the morphing model, the effect is with high reality, so this paper concerned in how accurate and fast to extracted face feature points from video sequence, then according to these characteristics, creating higher realistic 3d face model based on 3d morphing model.In order to accurately extracted face features, this paper proposed fusion skin color model and optical flow algorithm to detect face region, based on the summary of skin color model and optical flow. This method firstly with high clustering of YCgCr color space model, then the extraction pixel color of multi-resolution stratification optical flow algorithm, scanning candidate for accurate positioning face, then measured how to extract face feature points from the detected face region. In many face feature points extraction methods, this paper noticed based on geometric feature extraction method of feature point is simple, high efficiency. Finally, in the stage of model reconstruction, the topics include:(1) 3d face data acquisition and binarilized. 3d face data are the foundation of 3d face reconstruction, this paper mainly introduced the access method, and usually used 3d face database. Due to the 3d face registration is the key technology of construction 3d face morphable model, this paper introduced several registration method, and the method of this paper.(2) 3d facial feature extraction technology. This part mainly introduced the key feature points extracted in 3d model. Firstly, comparing the traditional methods for extracting 3d face features, then the improved method of extracting features AAM face with 3d face on the surface, based on the advantages and disadvantages of the traditional methods, was proposed.(3) Constructing 3d morphing model based on 3d original data, and extracting 2d face from video sequence to reconstruction 3d face model. Firstly, three-dimensional deformable model method and process were introduced, and then present the improved model matching algorithm: using stereo vision in establishing 3d morphing model to find the optimal model, namely to match with the minimum error 2d images of 3d model, then Using the maximum a posteriori probability to deduce the target function of the model error, finally, using the improved algorithm of Leven-Marquardt to optimize the objective function, deducing the matching parameters to matching the eventually 3d face model.The 3d model reconstruction system is developed using Visual Studio 2005 and OpenGL. Experimental results show that the video sequences for 2d images of 3d reconstruction based on 3d deformable model, the method of 3d face finally got rich texture, realistic models.