3D Face Modeling Based On Registering Of SICP

Traditional 3D face modeling methods have many deficiencies in modeling effects, whether can achieve real-time aspect and the computational complexity.3D face modeling based on Kinect is one of the best modeling methods at present, which is based on the principle of structured light and costs cheap, so the general users can use the device to build a face model. Compared with other modeling methods, this method has better performance in modeling results, the performance of real-time and so on. In this paper, the research work was carried out on 3D face modeling based on Kinect method and proposed appropriate solutions to corresponding flaws. The main works of this thesis and innovations are summarized as follows:Firstly, we first overviewd the background, significance and the situation of 3D face modeling, then gave a classified summary of existing methods of 3D face modeling and described all of the methods that were used in active modeling technique and passive modeling technique in detail. Sparse Iterative Closest Point was mainly studied in 3D face modeling.Secondly, as depth map obtained by the Kinect contains a lot of noise and lacks much information, combined with the traditional iterative closest point algorithm and improved iterative closest point algorithms can't properly deal with these issues, which will result in low registration accuracy, thereby affecting the quality of face model. This paper used sparse Iterative Closest Point to register the vertex maps. Sparse Iterative Closest Point algorithm could handle these weaknesses by formulating the registration optimization using sparse inducing norm.Lastly, In order to achieve real-time 3D face modeling, this paper researched a fast head segmentation algorithm which could quickly segment the user's head. Next, because the depth map obtained by the Kinect contains a lot of noise, this paper adopt use bilateral filter to optimize the depth image. Finally, the experiment results that used the method to build a 3D face model demonstrated the effectiveness of the algorithm.