Research On Iris Recognition Based On Deep Learning

Iris recognition technology is a biometric recognition technology widely used in identity verification.Due to the uniqueness,accuracy and security of iris,iris recognition has broad application prospects and scientific research value.With the development of electronic equipment,deep learning is also widely used in the field of image processing.From the core technology of iris recognition,a complete iris recognition system includes acquiring iris images,preprocessing iris images,feature extraction and feature matching.The focus of this paper is to complete the feature extraction and classification of iris combined with deep learning.The difficulty of iris feature extraction and classification mainly lies in that the iris image preprocessing is difficult first,such as eyelashes,eyelids,and glasses occlusion,and the influence of ambient light collected by iris directly affects the preprocessing results;secondly,there is a problem that the feature extraction method is handmade by experts.In view of these difficulties,this paper first analyzes the current research status of iris recognition technology at home and abroad,studies the application of convolutional neural network technology in traditional iris recognition technology in deep learning,and improves the performance of iris recognition technology.Secondly,the powerful feature extraction capability of convolutional neural network is combined with eye recognition to further optimize traditional iris recognition.The specific research contents of this article are as follows:(1)Position and segment the iris samples in the iris data set,and normalize the segmented iris to obtain rectangular iris samples.Using the powerful feature extraction capability of convolutional neural network,the Inception V3 network model is applied to the feature extraction link,and the normalized iris image data is used to train the network model.Finally,the classification and recognition of iris are completed.The recognition accuracy of the model in the test set is 98%.Although this scheme can obtain a high recognition rate,it still requires strict iris image preprocessing.(2)Aiming at the problem of pre-processing,this paper proposes an eye recognition network model based on Inception V3 network.In contrast to the iris,the entire eye circumference consists of eyelashes,pupils,iris,and sclera.Direct input of iris information into the network model can solve the problem of iris image preprocessing.At the same time,as an important supplement of iris recognition,it can make up for the problems caused by iris image failure in non-cooperative situations.The recognition accuracy in the test set was 98.5%.The experimental results show that the scheme is feasible,but considering the application in mobile terminal,the memory of the trained network model is too large.(3)Aiming at the problem that the training model is too large,this paper finally proposes an iris eye circumference recognition scheme based on the lightweight model.The size of the model trained by the lightweight model is 24184 KB,which is 3.8 times smaller than the model trained by the traditional convolutional neural network.At the same time,the recognition accuracy in the test set was 98.2%.The scheme not only reduces the volume of the model but also ensures the accuracy of iris recognition.Finally,the scheme proposed in this paper is compared with other iris recognition methods.While the recognition rate is guaranteed,it also has good generalization ability and robustness,especially the recognition of image samples from different angles.It has good research and application value.