Research On Multimodal Biometrics On Quaternion

Identity authentication is the problem people must face in daily life,for example,take money from the bank need six number password to authenticate,log in social software need account and password,all of this belong to traditional identity authentication that we should remember it by myself.It's easy to forget and it's unsafe.In today's daily life,in addition to the traditional identity authentication,biological recognition technology has also been widely used as a new identity authentication method.For example fingerprint unlock for smart phone,face log in for some software,iris recognition for access control system and etc.Compare with the traditional identity authentication,biological recognition is safer and more complicated,and we needn't to remember it.However,with the develop of science and technology,many kinds of camouflage technology appeared,single modal biological recognition has been unable to secure identity.Everyone needs the electronic commerce in this social,so the identity authentication is very important.Under such background,multimodal biometric technology arises at the historic moment.This article in view of the present multimodal biometric technology can just fuse less modal and the recognition rate is low,explore a new way of fusion,research the solution of multimodal fusion,the suitability of the classic pattern recognition algorithms,and the superiority of fusion algorithm.Proposed many kinds of multimodal biometric systems based on quaternion.And verify theses proposed algorithms have higher recognition rate than relevant algorithms through the simulation results.In this article,we proposed Quaternion PCA and Quaternion FDA in the third chapter,these two algorithms are based on linear transformation,the high-dimensional space projection data to low dimensional space.The core idea of quaternion principal component analysis algorithm is K-L transformation,and the quaternion linear discriminant analysis algorithm mainly borrowed from Fisher criterion function.We proposed Quaternion KPCA and Quaternion KFDA in the fourth chapter,on the basis of the previous chapter,introduce the concept of kernel function to extract the nonlinear feature of the multimodal biometric.And proposed Quaternion LPP and Quaternion DLPP in the fifth chapter.These two algorithms are based on manifold learning,in which the quaternion local keep projection algorithm based on Laplace feature mapping,it is the linear approximation of Laplace mapping the nonlinear feature extraction method.The other based on the previous and combine the Fisher discriminant analysis algorithm,introduced the concept of divergence in class and divergence between classes.In the end of each chapter,we list the simulation results of multimodal biometric systems with the proposed algorithm.The results show that the proposed six kinds of multimodal fusion algorithm all have the highest recognition rate compare with these single modal,series solution and parallel solution.Using biometric identity authentication is the trend of the development of science and technology,in view of the present means of attack mainly disguised attack,one algorithm that can verify a variety of biological characteristics and at the same time can guarantee the high accuracy is promising and valuable.