Extracting3D Human Facial Features Based On B-Spline Surface

Extracting3D facial features can be applied to the field of3D face reconstrction aswell as to the field of3D face recognition.Currently, the method of3D facereconstrction can be classified into3categories：method based on stereo vision, methodbased on orthogonal images, method based on an image and a single3D reference facemodel. In the method of3D face reconstruction using an image and a single referenceface model, the key step is to align the2D facial feature points in the image and the3dfacial feature points in the model, including extracting the facial features of the model.Firstly,we construct a3D face surface that has spatial struture information byfitting the scattered data points of the3D face model with B-Spline function. Based onthe knowledge of curvature,we extract the facial features’ region and their boundarypoints in the3D face model using prior knowledge of human face(including positions offacial features, concavity and convexity of human face surface). Here is the main workin this paper:1、Reading the PLY format file. Firstly,we analyze the structure of the PLY file,including the file header and the element data list.The file header defines the3D model’selements and its attributes. The data in the element data list is stored according to thesequence of the definitions of elements and attributes. Secondly, we analyze thestructure of the PLY file used in this paper. Lastly, we use OpenGL and VS2010to buildplatform to display3D data in the PLY file.2、Using B-Spline surface to fit the scattered data points. The information obtainedfrom PLY file is some scattered data points that don’t have any spatial structureinformation. In this paper, we use B-Spline to fit the scattered data points. Since theerror between the original data points and the surface in which B-Spline fitting is usedonly once is large, we decide to adopt multilevel B-Spline to maintain the goodsmoothness and improve the approximation accuracy of the surface.3、Extraction of3D facial feature. We first introduce the basic knowledge ofcurvature, perform an analysis of various kinds of curvatures and make a comparisonthrough calculating the curvatures.Secondly, the average curvature is selected to carve up concave and convex region on the3D face surface, and then facial feature regionsare extracted.Finally, taking the prior knowledge of human face as the constraints, weuse the convexity and extremum properties of a surface’s cross-section curves to extractboundary points of the facial feature areas and the geometric parameters of a featurearea are given.