Research On Discriminative Biometrics Feature Extraction And Key Generation

The rapid development of science and technology makes information security becoming the most challenging issue in the process of global informatization. Ensuring data security is the most important issue worthy of consideration. In traditional cryptography, the security of the system is totally dependent on key security. Because keys are lack of necessary link with users, the system can not distinguish the identity of the key users and can not determine the key user is an authorized user or a malicious attacker, resulting in illegal key sharing. Biometric key are generated from the user's unique biometric features and can effectively solve the traditional security problems in cryptography.Biometrics based key technology includes three aspects:feature extraction, biometric key generation and security design. Biometric feature extraction is the primary step of biometric key generation. Security Designs are aimed at security vulnerabilities in biometric key generation system and protect the user's biometric information. This paper mainly focuses on those three aspects, the main innovations are as follows:In chapter 1, a systematic analysis of biometrics feature extraction, biometrics key generation, biometrics algorithm evaluation has been described.In chapter 2, to address the limitation of UDP, maximum variance difference embedding is derived from maximizing difference between global variance and local variance, it utilizes the maximum variance difference criterion rather than the generalized Rayleigh quotient as a class separability measure, thereby avoiding the singularity problem. Besides, the projection space of maximum variance difference embedding is an orthogonal subspace.In chapter 3, Maximum margin neighborhood preserving embedding algorithm modifies the neighborhood preserving embedding by maximizing the maximum margin distance while preserving the geometric structure of manifold. Besides, by embedding the objective of modified maximum margin criterion into the objective of LPP, maximum margin locality preserving projection was constructed which largerly enhances the performance of LPP.In chapter 4, Discriminant locality preserving projection (DLPP) encodes discriminant information into the objective of locality preserving projection and improves its classification ability. However, DLPP can not obtain optimal discriminant vectors which utmostly optimize the objective of DLPP. Our proposed direct discriminant locality preserving projections(DDLPP) algorithm directly optimizes discriminant locality preserving criterion on high-dimensional vector space via simultaneous diagonalization, without any dimensionality reduction preprocessing. We have also analyzed polynomial expansion and Gabor filters based DDLPP algorithms.In chapter 5, Gradientfaces only takes two mutual vertical directions into account, which may not be adequate for face description. As different directional gradients can reflect different face feature, we consider multiple directions and construct a multi-directional orthogonal gradient phase face (OGPF) algorithm by introducing directional derivative into image calculation, which can provide more complete face feature description, In our multi-directional OGPF algorithm, multiple OGPF can be generated from one image, which extends the samples of each person and is beneficial to many subspace based dimensionality reduction techniques.In chapter 6, several works have been done to address the problems in biometrics based key generation, which include:1) proposed a new method for biometric vector quantization; 2) Proposed a cancelable biometric key generation scheme by utilizing chaotic random projection; 3)constructed a biometrics based key generation system by utilizing the orientation feature of finger vein and fuzzy commitment; 4) utilize the minutial feature of finger vein and fuzzy vault to construct a biometrics based key generation system.