Research Of 3D Face Recognition Under Expression Variations

Face recognition technology becomes one of the most effective means of identity authentication in different situations as it has high practicability and reliability.The technology is mainly divided into two categories: 2D(two-dimensional)face recognition and 3D(three-dimensional)face recognition.Compared with traditional 2D face image,3D face data is a direct representation of real human face in three-dimensional space.It can more comprehensively describe the shape and topological structure of the face,and is not affected by illumination,pose or makeup.However,expression variation is a huge challenge for 3D face recognition.In order to solve the problems and improve the recognition performance under expression variations,the main work and innovations are proposed as follows:1.A 3D face recognition algorithm using keypoints and local shape descriptors is proposed.Firstly,after preprocessing,multiscale shape variation index is calculated to locate keypoints on the 3D face.Secondly,the 3D histograms of normal distributions(3DHoND)descriptors are constructed at the keypoints and a large number of irrelevant gallery faces are eliminated based on the descriptors.Thirdly,the keypoints of the remaining faces are matched based on the covariance matrix descriptors generated as local shape descriptors.Finally,the similarity of two faces is measured by the number of the keypoints that can be correctly matched.The algorithm can detect the keypoints that are stable under expression variations and extract a variety of geometric information,which effectively reduces the impact of expression variations.The experiments of the proposed algorithm are carried out on the Bosphorus,FRGC v2.0 and BU-3DFE datasets and achieve superior recognition performance.The results demonstrate that the proposed algorithm is robust under expression variations and outperforms the state-of-the-art algorithms in term of the recognition speed.2.A novel algorithm for 3D face recognition based on depth map of shape index and iso-geodesic facial curves is proposed.Firstly,depth maps of shape index are generated on XOY planes based on the face after preprocessing,and LSDP(Local Shift Derivative Pattern)features are extracted from the depth maps of shape index after removing the mouth area and matched.At the same time,a set of iso-geodesic facial curves are extracted and sampled from the face and matched by improved Hausdorff distance.Finally,classification is carried out by fusing the matching results of depth maps of shape index and iso-geodesic facial curves.The algorithm extracts quite stable features under expression variations and combines the advantages of global and local features.As the results of experiments on FRGC v2.0 and Bosphorus datasets show,the proposed algorithm not only has better recognition accuracy,but also has better robustness under expression variations.