Research Of 3D Face Recognition Based On Geometric Feature Vector

At present, research of 3D face recognition has become one of hot issues in academic research. This thesis studies 3D face recognition method based on geometric feature vector: first of all, read the 3D face point-cloud data to extract the facial contour, and then locate some key feature points and calculate the geometric feature vectors. Finally we get recognition results by calculating the similarity. In this thesis, the main works are carried out:(1) Data Preprocessing: Extract the point-cloud data from 3D face data files of 3DS format at first. After determining the 3D facial coordinate system, extract middle facial contour and transverse contour which goes over the apex of nose combined with the depth information. And then we try to smooth and fit the contour that we extracted before. This process reduces disturbance from the data noise, and improves the accuracy of point positioning in the next step.(2) Key Feature Points Positioning and Eigenvector Group Calculation: Locate and extract the 11 key feature points of human face by calculating the curvature, then select and calculate 21-dimensional feature vector of 6 major categories, (including Distance, Angle, Area, Circumference, Volume and Face ratio feature), constituting a relatively complete set of geometric feature vector group to identify matching work. In this process, we put forward the concept named the "Triangle Skeleton of Human Face" to represent the abstract 3D human face.(3) Similarity Measure and Face Recognition: After analyzing the feature vector group, we strike a balance between stability and discrimination, give each dimension vector different weights, and then process them with normalization method. In the end, we complete the recognition by calculating the similarity between Identification Database and Sample Database. In the course of the entire algorithm, we use 5 calculation method (including the neighborhood search method, the average method, the symmetry correction method, stratification methods and abstract methods) comprehensively to try to improve the effectiveness and robustness of the recognition algorithm. (4) 3D Face Recognition System and Recognition Rate: we build a visualized 3D face recognition system based on OpenGL, Visual C++ and SQL database technology. This system can extract facial contour, locate key facial feature points, calculate the feature vector group of human face, and complete the recognition work by comparing similarity. Based on this system, we choose 42 test data to do the recognition test on 150 samples in the database (got recognition rate for 90.5%, error rate for 9.5%), and relevant data are compared and analyzed in the end.