Research On Face Recognition Using Geodesic Distance

Face recognition is an important research direction in biometric category. Because the three-dimensional human face possesses six degrees of freedom, and it contains more data information, so researchers have gradually begun to withdraw their eyes from the two-dimensional face research turned to the three-dimensional face recognition research in modern times. Geodesic distance is a concept of mathematical morphology, compared to the traditional distance measurement, it is better to overcome the effects of changes in facial expression and posture changes caused by other issues, in view of the above two advantages, this paper preliminarily explored the two-dimensional faces and three-dimensional face recognition methods on the basis of measure distance theory, the paper's main research contents and results are as follows:This paper proposed a method of the KPCA face recognition of geodesic distance, which was aimed at the problem that the information in the face detection data are high eigenvector and face recognition is easily affected by expression changes, where, the principal component was extracted with the nonlinear method. First, KPCA method was used, the face data were mapped to the high dimensional space, and then the principal component of the face was extracted in high dimensional space, where, polynomial kernel was chosen as kernel function, finally, geodesic distance was introduced to replace the original Euclidean as similarity measure, which could more accurately measure the actual distance between two points, and face recognition rate was affected less by the expression changes. The method could not only achieve dimension reduction, but also achieve the purpose of extracting the data effectively. At the same time, the robustness of identification is strong, and recognition rate is higher.This paper proposed a three-dimensional face location algorithm based on the geodesic distance, which was aimed at the problem that face location accuracy of traditional method is lower. Algorithm of three-dimensiona face localization was improved. Geodesic distance could better overcome the effects caused by posture and expression changes. For a given three-dimensional human face, first, the methods of filtering was carried out to process basic data for the face, and then a feature nose point was lacated on the basis of the processed data, and then fixed the whole face, finally, the face model unified under the same framework to coordinate in order to conduct the follow-up to people face feature extraction and matching, results show that the effectiveness of this method.This paper proposed face recognition of feature matching based on geodesic distance, where was aimed at the problem that complex face data lead to mismatch and difficultly to identify. 3D face data information and the advantages of 2D face algorithms were combined,and experimental analysis was carried out on acquired 2D face virtual image, used SIFT method to extract feature, introduced random projection in order to reduce the amount of calculation, and combined with sparse representation method to classify, finally, introduced the concept of geodesic distance and Angle to assist face matching, this reduced the difficulty to match and also it guaranteed the correctness of matching and improved the final effects of identification and robustness.