Studies On Key Issues Of3D Face Recognition

With the development of science and technology, traditional password authenticationsystem can not meet the needs in modern society. So more natural, convenient andeffective identification technology and system are urgently needed. The biometricidentification technology, which utilizes body’s immanent physiological characteristicsand behavioral characteristics, is becoming an important and prospective field of theidentification technology with its merits: not easy to change, difficult to counterfeit, easyto carry, etc. Because of the tremendous value of theoretical significance and practicalapplication,3D face recognition technology as a typical biometric identificationtechnology has become an important research direction in image processing, computervision, pattern recognition and artificial intelligence.3D face recognition technology hasmany merits: non-contact, complete face information, insensitive to light conditions, etc.But it also restricted by data processing difficulties, pose, expression and age factors andlow-level automation.Based on the latest status of domestic and foreign research in the field of3D facerecognition, this dissertation studies intensively several issues of3D face recognition.Through theoretical analysis and experiments, the validity of the proposed algorithm isverified in this dissertation. The innovations and contributions are as follows:(1)Traditional face recognition methods only use2D gray information from faceimages. This dissertation extends eigenface and fisherface into the complex domain andcombines2D gray data with3D depth maps. Experiments show that the proposedalgorithms after information fusion outperform the traditional eigenface and fisherface.(2) It is difficult to get a satisfied recognition rate with only2D gray informationunder different poses.3D face data contains original facial geometric information.Especially3D face depth maps have geometry consistency after rotation. Face images indifferent poses are corrected into neutral front face by rotation and interpolation. The mainpoint of this project is to make full use of3D information of face images and reduce theimpact of different poses.(3) With the development of imaging technology, image resolution is becominghigher and the amount of3D images data is larger. This dissertation proposes a facefeature extraction algorithm based on contour theory. The features of3D face depth mapsare extracted by surface contour and Fourier descriptors and then face recognition is basedon these contour features. The experiments with face database obtain good recognition accurancy.(4) In recent years, manifold learning has been widely applied in the fields of patternrecognition, data mining, machine learning and computer vision research. Manifoldlearning is a branch of nonlinear dimension reduction. Isomap and LLE are tworepresentative methods. This study introduces manifold learning into dimension reductionprocess of3D face images and establishes the dimension reduction statistics model of3Dface images. The face recognition results with face database show that this novel method ispotential for data dimension reduction.