Research On Bimodal Biometric Recognition Algorithm Based On Manifold Learning

With the rapid development of computer science and information technology,Humans has entered the era of Internet and Big Data.People pay more attention to the security of personal information than ever before.It is extremely important to confirm the authenticity of identity information between the information exchangers,when a large amount of information is exchanged.As a kind of authentication method,biological characteristics have been widely used with the advantages of the natural,intuitive,safe and fast and can balance the contradiction between system security and user experience.Identifying these biological characteristics with a machine has always been a difficult field because of the complex structure of the biometric features,the diversity of forms,and the volatility of the acquisition process.In recent years,one of the hot issues in the field of machine learning is how to find the characterization of low-dimensional space embedded in high-dimensional space.Manifold learning happens to have this feature to be a good solution to the problem of biometric extraction and dimension reduction,which is widely used in the field of biometrics.The existing biometric authentication system mostly certifies a single feature,and the system has its own shortcomings and limitations of use,and can't be widely used in special people.Based on the above reasons,the identity authentication system,which can identify a variety of biometric simultaneously,will have a better use and application value,and get the researchers' attention in the field.In the third chapter of this paper,the extended LPP algorithm,the extended DLPP algorithm and the extended ILPP algorithm are proposed for the shortcomings of the traditional LPP algorithm.In the classification process,the between class scatter diagram is introduced by extended DLL algorithm to improve the recognition rate by taking into account the between classes discriminant information.The extended ILPPalgorithm transforms the LPP algorithm into orthogonal projection mapping by Schmitt's orthogonal transformation,and can effectively remove the orthogonal information that affects the recognition rate in the image extraction process.In the fourth chapter,we propose an extended locality fisher discriminant analysiand an extended marginal fisher analysis by combined Fisher discriminant analysis with the manifold learning method.The extended locality fisher discriminant analysiand algorithm makes it possible to have the advantages of linear discriminant analysis and manifold learning by using intraclass divisiveness and interclass dispersion matrices.The extended marginal fisher analysis algorithm successfully separates the margin between the distant points in the same classes and the closer points in the different classes by the penalty graph.At the end of each section,the proposed algorithm is applied to the dual-mode biometric system to verify the effectiveness of the algorithm.