Research On Iris Recognition Algorithm Based On Deep Learning

Information technology give wings of convenience to people's life,but then there is the disclosure of personal privacy and other issues,people attach increasing importance to the protection of privacy.iris are internal tissues that do not wear out and are difficult to forge,making them reliable and effective biometric feature.The biometric fusion method can make use of multiple biometric information to further improve the accuracy of the identification system.Deep learning has a very good performance in classification,similarly deep learning in biometric recognition is also rapid develop,so the study of iris recognition based on deep learning is of great significance.In the aspect of iris recognition based on deep learning,this part consists of two parts,the first part is to study the influence of data quantity on iris recognition based on deep learning.The second part is to study the effect of data pre-processing on iris recognition based on deep learning,using two kinds of normalization methods,scheme one:polar coordinates unfold,scheme two:iris weighted filling of pupils,and using histogram equilibrium and Laplace transformation to enhance the image after normalization,respectively;Different schemes were studied on different networks,and the recognition rate of VGG16 networks reached 99.520%.In terms of binocular iris fusion recognition,the binocular iris fusion recognition algorithm in this paper has three steps:The first step is the training of the iris feature extraction network for the left eye subclass and the right eye subclass.From the data set,select categories where the left eye subclass and the right eye subclass are not empty sets,and then expand the data and divide it into the training set and the verification set,then perform iris weighted filling and train VGG-16,and the trained network is used as a feature extraction network;The second step is to select the training set in the data set again based on the small number of left-eye sub-categories and right-eye sub-categories in the data set,the left-eye sub-category and the right-eye sub-category of the selected data set are extracted by the feature extraction network to a one-dimensional feature vector with a length of 12800,and the two are connected in parallel,and uses PC A or LLE to reduce the dimensionality of the paralleled features;in the third step,the dimensionality-reduced data set is placed in the SVM for training.Experiments have proved that when the quality of a certain iris image collected is not ideal when performing identity authentication,the fusion of the irises of both eyes can improve the accuracy of recognition.