Research On 3D Facial Surface Construction And Feature Extraction Based On B-splines

The fundamental principle of computer stereo vision with dual-camera, that is, the binocular vision, has a strong resemblance to the visual perception procedure of humankind. Getting different images via two different viewpoints and then calculating the position deviation of image pixels based on triangular measure, and then, the depth information of the object that we interested can be attained. After obtaining depth information of 3D face, we construct a 3D facial model with better expression by using reasonable spatial structure, which is the procedure of 3D facial surface construction. By the 3D facial model we can extract some important 3D facial features which will be used for face recognition.In this thesis, we do further research on 3D facial surface construction and feature extraction based on B-splines. Some novel algorithms are presented, our work contributes to the following aspects.1,Obtaining 3D facial data. We introduce the principle of binocular vision, and compare several methods of 3D facial data acquisition. Then, we can obtain the 3D facial data by two 2D images which were taken with two parallel cameras. In addition, we also analyze and obtain data for following research from 3D facial databases provided by other research institutions.2,Constructing 3D facial surface. Considering these data are scattered points without structural information, we present a new algorithm for surface fitting using multilevel B-splines approximation, after that we can construct 3D facial surface with C~2 continuous. In order to reduce the deviation in surface fitting, we improve the algorithm by making the scattered data uniform before multilevel B-splines approximation. The experiment proves that the improved algorithm can reduce deviation and the surface will coincide more closely with the data.3,Extracting 3D facial features. First, we analyze a variety of surface curvatures and calculate these curvatures of the 3D facial surface. Then, we present a method of 3D facial feature extraction based on curvature information. By using the method we can extract some important facial features. Further, these features will be used for 3D face recognition.