Research On Three Dimensional Face Recognition And Template Protection

Identity authentication is the most important basis of information security system.Compared with passwords and smart cards, biometric identification is unforgettable andnot easy to lose, which affords users higher level security protection. Thus, biometricidentification has become the trend for identity authentication.Nowadays, the security vulnerability of existing biometric identification technologyhas drawn more and more attention of researchers. Most of the security threats are causedby traditional direct storage templates.he technology of biometric template protection hasemerged in recent years, which turns the traditional template into a secret style to store.As such, there will be no information leakage of the original templates during identityauthentication. However, there still exist many theoretical and application problems.While biometric template protection technology ensures the security of templates, itbrings great challenges to existing identification technologies. As the template informa-tion is not directly stored, most existing 3D face recognition algorithms cannot work dueto the restrictions for registration, feature extraction and feature matching.In this thesis, we adopt the template protection technology to 3D face recognitionunder restrictions. According to different styles of 3D face templates, we proposed twotemplate protection algorithms based on hamming-distance space and sequence-distancespace respectively. We analyze the theoretical and application security of the algorithmsin detail. The contributions of our thesis are as follows:We have proposed a 3D face preprocessing method for template protection. Ourmethod could preprocess information efficiently, accurately, and automatically. Ourmethod works on the original 3D face points-cloud model without topological informa-tion, which isolates the head from shoulder by analyzing the 3D data distribution. Com-bined with the accurate location of some basic feature points and distribution of five senseorgans, the face area is accurately segmented. At last, the face pose is corrected accordingto the three projection angles of the nose bridge on the three facets. The effectiveness ofour preprocessing method is shown through the simulation results.We have proposed two typically feature extraction algorithms for template protec-tion. The first algorithm is based on local statistic features, containing the histogram values of point number and energy that are extracted from the divided horizontal layer.The second one is based on global features using space integration. We resample the 3Dface model and integrate the gray value, curve rate and depth value into a new feature, anduse pattern recognition to train the feature. The two algorithms are proved to be highlyeffective for recognition.As for the long feature templates, we have designed a template protection algorithmbased on key-generating-SMS (self-masked scheme) and applied it in the feature extrac-tion algorithm based on local statistic feature template protection. We analyze the securityof our algorithm in the view of information theory and propose a comprehensive analy-sis method of security and recognition performance. Finally, we provide a quantitativeanalysis of the algorithm in specific application with experiment results.As for the long feature templates, we have designed a template protection algorithmbased on key-binding-SST (secret share template ). Secret sharing technology is adoptedand double error tolerances mechanisms are designed to improve the error tolerancescapacity in SST. High security level and recognition performance can be achieved at thecost of efficiency in SST. We applied the algorithm in the feature extraction algorithmbased on global features using space integration. The result showed that SST scheme isthe best template protection algorithm.