Research On Iris Image Detection Based On Convolution And Recurrent Neural Network

A complete iris recognition system consists of six steps: iris image acquisition,iris image quality evaluation,iris image correlation preprocessing,iris region location,iris feature extraction and iris feature recognition.The accuracy of iris recognition system is not only related to recognition algorithm,but also to the quality of input image.The quality of the input image is determined by the filtering accuracy of the original image.High quality iris image not only needs clear and complete iris texture,but also ensures its authenticity.The traditional iris original acquisition image detection process commonly uses the method of manual screening to build the initial data set,and then according to the relevant evaluation methods,preprocessing methods,respectively,to process the initial data set,and finally form the available iris data set.However,such a method has a large amount of human intervention,and the work efficiency of separation is relatively low.If it aims at a large number of data tasks,it will form a great deal of processing pressure.Therefore,in order to solve the efficiency problems caused by human screening intervention and separate operation,this paper combines the first three tasks of iris recognition system process to separate out the out of focus and occluded original iris images,and judges the authenticity of the original iris image collection sequence,this paper proposes a detection process for iris original collection image in order to ensure the accuracy and effectiveness of iris identification work in the later stage of the recognition system.At the same time,on the basis of the detection process,combined with the neural network model,this paper proposes a classification detection model based on convolution neural network and recurrent neural network.This model is mainly used to classify and judge the availability and authenticity of iris images,and efficiently and batch screening the original collection data.The experimental results show that in iris image usability classification,the AUC value of conv1,the best structure layer in vgg16 network structure,can reach 0.88,and the average accuracy of classification on test set can reach 99%,the results show that the structure of iris usability classification proposed in this paper can classify the original iris data in batches and accurately,and can replace the problem of manual selection of the original iris image,which makes the original iris image selection more reasonable;In iris image authenticity classification,the highest accuracy of the test set under the pre trained conv1 structure layer is 58%.However,the test trend still shows an upward trend under the fixed epoch settings and limited data changes in this experiment,the results show that the structure of iris authenticity classification proposed in this paper can be used for reference in the research and expansion of iris image security field,so as to provide more meaningful data base for subsequent iris image recognition from the source.The main work and contributions of this paper are as follows:(1)According to the common operation flow of iris recognition system,this paper puts forward a research flow of iris detection based on the existing problems and shortcomings of these steps in the actual operation process.This flow is dedicated to the integration and implementation of the separated steps In order to improve the overall efficiency and accuracy of iris recognition system,the availability and authenticity of the collected imageare classified and detected.(2)In the classification of iris image usability,because the original collected image belongs to the unlabeled image,in order to realize the unlabeled image classification of neural network.In this paper,vgg16 model structure in convolutional neural network is selected.According to the standard of iris image quality evaluation and K-means method in clustering method,iris image usability classification is carried out for the unlabeled original collected image.By comparing the pre training of vgg16 network structure and the effect of different convolution layers,the results are explained and demonstrated.(3)In the iris image authenticity classification,considering the serialization characteristics in the process of image acquisition,the LSTM network structure in the cyclic neural network is adopted.Based on the convolution neural network structure trained in the iris image usability classification,the serialized image features are extracted and the iris sequence acquired image authenticity classification is carried out.In iris image usability classification,we choose the structure with the best effect under different levels.On this basis,we continue to carry out the comparative experiment of this category,and explain the results.