Research on Algorithms of 2D Face Template Protection
|School||Harbin Institute of Technology|
|Course||Instrument Science and Technology|
|Keywords||Face Recognition Template Protection Non-Uniform Quantization Help Data System|
Biometric recognition offers a reliable and natural solution to the problem of user authentication in identity management systems. With the widespread deployment of biometric systems in various applications, there are increasing concerns about the security and privacy of biometric technology.Template protection aims to resolve the security problems in traditional biometric system by transforming the biometrics into an encryption form. It protects the users’privacy, makes the templates reusable and improves the system security.2D face recognition is one of the hot subjects in biometrics study, and is widely used in biometrics for its series of advantages. Therefore, the study of 2D face template protection technology is very important and with significant practical meanings.At present, statistic-based face recognition algorithm becomes a leading algorithm in 2D face recognition research, for its simplicity and effectiveness. However, feature vectors extracted by statistic-based face recognition algorithm have few components. This simplifies the computation during the match process, but brings some disadvantages for template protection at the same time. Therefore, this paper will mainly research template protection algorithm based on statistic-based face recognition algorithm.Firstly the paper summarized the algorithms of template protection, and introduced the attacks of traditional biometrics system as well as the basic scheme of template protection system with cryptology. Key methods in the scheme, such as Hash function, error tolerant technology etc., were descried in the end.Secondly, the paper utilized three typical kinds of statistic-based face recognition algorithm to extract face template from 2D face images. They are Eigenface, Fisherface and Commom Vector. Besides, the paper improved a nonlinear face recognition algorithm Complete Kernel Fisher Discriminant (CKFD) to get a new algorithm, named CFD. These four recognition algorithm were tested to get the performance comparison and template characteristics analysis. It provided the reference for the next step for the template protection algorithm design. Thirdly, based on HDS proposed by Philip research, the paper proposed a new template protection algorithm. Several key steps in HDS were modified to suit statistic-based face recognition. They are, feature extension, multi-threshold quantization etc. Four recognition algorithms were applied to the proposed template protection scheme. Test results showed that the effectiveness of the algorithm and feasibility of scheme. Besides, the security analysis demonstrated the security of the proposed algorithm.Finally, based on non-uniform quantization, this paper proposed another face template protection scheme. The extracted feature vectors were quantized by non-uniform method and binding with a random key. It improved the system security a lot. Four face recognition algorithms were applied to the proposed scheme. Test results showed that the algorithm got good recognition results.