Research On Identification And Verification Technologies Of Cancelable Palmprint

As an emerging biometric, palmprint feature is convenient and reliable for personal identification and verification. However, with the application and popularization of palmprint technologies, the security problems of palmprint identification and verification systems gradually expose, which also exist in other biometric systems.●Biometrics are immutable and cannot be updated. A user cannot revoke and reissue his biometric template even if his biometric template is compromised.●A user’s biometric templates of the same feature likely are stored and shared in various databases with more and more biometric systems appearing in our daily life. If one user’s biometric template in one database is attacked successfully, the user’s biometric templates in the other databases are not secure anymore.●Biometrics likely leak users’private information, such as gene defects, diseases and so on.Everyone has only two palms, so the amount of palmprint features is very limited. The use of palmprint without any protection leads to serious security risks and privacy problems. Cancelable biometric is one of the most promising schemes for biometric protection, so this dissertation focuses on the identification and verification technologies of cancelable palmprint. The major contributions and innovative points are as following:(1) Dynamic Weighted Discrimination Power AnalysisThe framework of face and palmprint recognition algorithms based on discrete cosine transform (DCT) is summarized. The premasking and selection of DCT coefficients are intensively analyzed and reasonably explained. Dynamic weighted discrimination power analysis (DWDPA) avoids the complex optimization of the shape and size of the premasking window and achieves more efficient coefficient selection. Normalization avoids the discriminant interference among DCT coefficients. Dynamic weighting enhances the discriminant ability of DCT coefficients. The weights are adjusted adaptively according to various databases, so DWDPA is more universal and robust to various databases. Furthermore, cancelable biometric schemes are developed and implemented based on DWDPA.(2) Two-dimensional Cancelable Biometric SchemesRandom compressive projection (RCP) is extended to two-dimensional RCP (2DRCP) and two-directional two-dimensional RCP ((2D)2RCP), so the storage and computational costs are both reduced. Fusing2DRCP and other two-dimensional dimensionality reduction technologies, two-directional two-dimensional fusion RCP ((2D)2FRCP) achieves complementary advantages.(2D)2FRCP methods have no ideal performance when all users have different tokens, so two-directional two-dimensional plus RP ((2D)2PRCP) is proposed to avoid the drawbacks of (2D)2FRCP. Comparing different combinations, feasible two-dimensional cancelable biometric schemes are determined.(3)2DPalmHash CodePalmHash Code based on texture feature is proposed for secure palmprint verification. One-dimensional (1D) PalmHash Code is extended to two-dimensional (2D) PalmHash Code, so the storage and computational costs are both reduced. In2DPalmHash Code, two-dimensional random projection (2DRP) of Gabor-filtered palmprint images is utilized as feature extractor to generate noise-like cancelable templates. Since2DRP directly leverages the matrix structure, the accuracy is high. Using the empirical distribution of real part and imaginary part of Gabor-filtered palmprint images, which approximately obeys t location scale distribution, the verification performance of PalmHash Code is systematically analyzed.(4)2DPalmPhasor CodePalmPhasor Code, which employs non-linear operating instead of linear operating, is proposed to enhance the security of cancelable palmprint codes.1DPalmPhasor Code is extended to2DPalmPhasor Code, so the storage and computational costs are both reduced. Moreover, the verification accuracy is improved. Based on the distribution character of real part and imaginary part of Gabor-filtered palmprint images, the verification performance of PalmPhasor Code is systematically analyzed.(5) Orientation Co-occurrence Cancelable Palmprint CodeThe framework of orientation co-occurrence cancelable palmprint code is proposed. Multi-orientation score level fusion reduces the variance of matching score and improves the verification performance. Vertical orientation transposition is used to scatter the high correlation of two-dimensional cancelable palmprint codes along vertical orientation and enhance the robustness against statistical attacks. A complete analytic framework and experiments confirm that orientation co-occurrence two-dimensional cancelable palmprint codes, which are enhanced by the two aforementioned measures, satisfactorily meet the objectives of cancelable biometric at the same time. Even when users’tokens are stolen, the accuracies of orientation co-occurrence two-dimensional cancelable palmprint codes are still high.(6) Dual-key-binding Cancelable Palmprint CryptosystemTraditional biometric key extraction methods do not consider the cancelable ability of biometric key. Once the biometric key is exposed, the users, whose biometric keys are stolen, are never able to use the biometric-based authentication systems or cryptosystems anymore. Base on the proposed two-dimensional cancelable palmprint codes, dual-key-binding cancelable palmprint cryptosystem is developed, in which the diversiform palmprint keys can be revoked and reissued. Dual-key-binding framework enhances non-invertibility of the system to effectively protect the original palmprint features; moreover, it enhances the robustness against select-plaintext attack.In summary, the algorithms and schemes proposed in this dissertation effectively enhance the security and privacy of palmprint, and also facilitate the public acceptance of palmprint systems. The corresponding achievements not only provide the identification and verification technologies of cancelable biometric with important theoretical reference, but also promote the standard establishment and application popularization of biometrics. Thus the contributions of this dissertation have obvious social benefits and economic values.