Biometric Identification Method Based On Multi-View Learning

With the development of modem social information and multimedia technology,the amount of data generated by the society is increasing rapidly,and a large number of data can be described in a variety of different views.For example,the application of face recognition,feature extraction can form different forms of the same sample face images.Compared with the classification method of single view and multi view learning methods attempt to remove the interrelated and complementary features in different views within,so it can improve the classification effect on data sets;in addition there are many learning methods based on single view can also be extended to the view layer,therefore,produce most of the inner nature of the map data the algorithm can be sufficient and varied information to explain things.In this paper,multi view learning method is used for biometric identification,do the following work:Firstly,based on the Fisher discriminant analysis inspired by single view case,the classical discriminant model is extended to make it suitable for multi view data.The main purpose is to rewrite the future unification from all views projection data to a common subspace in the subspace of similar data closer to each other,heterogeneous data mutual spread,to make full use of discriminant information between views,and then improve the classification effect;Secondly,on the basis of the first point,in order to further remove the redundant information in the original data,and better use of local structure information,the introduction of irrelevant information to improve the classification accuracy of local information technology;Thirdly,the classification of multi view data containing unlabeled data is discussed.Based on the first point,the multi view model is extended to semi supervised discriminant model,using a label like tectonic discrimination model,and based on the inspiration of manifold learning,the use of unlabeled samples to keep the local structure information of the original multi view data;Fourthly,the reality of multi view original data is often linearly inseparable,based on the first point on the kernel of high dimensional map of the original data,and the simplified calculation process of nuclear techniques,design a linear noulinear case discriminant model,to further improve the ability of classification model.Finally,this paper chooses the multi view data set was tested,including in handwritten data sets,multi spectral and color face data sets were compared using cross validation,the proposed algorithm and the existing multi view learning representative methods for qualitative analysis and experimental results prove the effectiveness of the algorithm the.