Research on Discriminative Biometrics Feature Extraction and Key Generation
|School||Southwest Jiaotong University|
|Course||Signal and Information Processing|
|Keywords||Information Security Biometric key Maximum variance differential embedded Maximum edge close neighbors to keep the projection Maximum edge locality preserving projection Direct identification locality preserving projection Multi-directional orthogonal gradient phase face Chaotic random projection Fuzzy commitment Fuzzy safes Finger vein|
The rapid development of science and technology makes information security has become the world's most challenging issues in the process of information, ensure the security of the data is the most important issue worth considering information security. Traditional cryptographic technology, the security of the system is completely dependent on the security of the key. Due to lack of necessary link between the key and the user, the system can not distinguish between the key the user's identity, that is not to judge the authorized user or a malicious attack, resulting in illegal secret sharing. The biometric key is generated from the user unique biometric key, can effectively solve the security issues in the conventional cryptography. The biometric key technology mainly includes three aspects: to identify biological feature extraction, biometric key generation and security design. Identify biological feature extraction biometric key generation, key generation and has important implications. Biometric key generation refers to the use of biometrics and cryptography algorithm to generate a stable sequence technology. Biometric security program designed for security vulnerabilities in biometric key generation system design to protect the key and the user's biometric information. This thesis is a study around these three areas, based on analysis of previous work, the main innovation of the work are as follows: 1) for the the UDP algorithm in the small sample, the maximum variance differential embedded algorithm (VDE), the algorithm directly obtained by solving an eigenvalue problem and the projection matrix without matrix inverse operation, so the small sample size problem VDE In addition, as the eigenvalue problem for the orthogonal eigenvectors. 2) local preserving projection (LPP) emphasis on the local characteristics of the data rather than directly for classification problems, in order to the LPP extract feature has the ability to identify the maximum margin criterion embedded locality preserving projection algorithm, maximum edge local holding projection algorithm (MMLPP), MMLPP enhance the ability to identify locality preserving projection algorithm feature extraction. The local linear discriminant embedding (LLDE) embedded into the maximum margin criterion of NPE in the objective function, thereby enhancing the NPE algorithm extracts a feature the ability to identify. However, to maximize data class LLDE difficult dispersion and minimize class dispersion and dispersion between the MMNPE algorithm directly maximize data class mentioned and minimize the within-class dispersion than obtained performance than LLDE better. 3) preserving projections (DLPP) for the identification of local small sample size problem, referred to directly identify locality preserving projection algorithm execution DLPP guidelines without matrix inversion, and without using PCA pre dimensionality reduction, but quadratic eigenvalue solution and a projection matrix, the process use the DLPP criteria to maximize extraction of the authentication information of the data. In addition, based on the polynomial expansion and Gabor filters DDLPP algorithm further improve the ability to identify the biometric template. 4) gradient Face merely describes a human face in the horizontal direction and the vertical direction of the gradient, and the experiments show that the performance of the algorithm of the gradient face is less stable. Orthogonal to the mentioned multi-directional gradient phase face algorithm plus data dimension of the subspace method improves the recognition accuracy of the algorithm of the gradient face, and enhance the stability of the gradient face algorithm. 5) for biometric key generation embarked on a series of work, including: 1) a new quantitative method; 2) a chaotic random projection and fuzzy commitment algorithm to generate cancelable biometrics feature key programs and simulation studies; 3) the use of finger vein texture features and fuzzy commitment algorithm to build a biometric key generation system; 4) the use of finger vein minutiae features and fuzzy safe algorithm to construct a biological characterized in the key generation system.