Research On Iris Image Gender Recognition Technology Based On Deep Learning

Biometrics-based identity authentication is a requirement for the development of information technology and economic globalization,and it is also one of the important technologies urgently required by the government and business fields.Among the biometrics,the structure of the iris is unique to the individual and remains stable with age.Authentication through iris is considered to be meaningful and promising.The earliest application of iris recognition as a personal identity can be traced back to the Paris penal system,which distinguishes prisoners by visually inspecting the prisoner's iris,especially the color of the iris.In recent years,more and more scholars have paid attention to iris recognition.This paper combines deep learning algorithm to realize gender prediction through iris images.The main research contents are as follows:In the collected original iris image,there are occlusions of the iris area such as eyelashes and eyelids.In order to reduce the interference information when extracting iris features,the iris image is first segmented.This paper proposes a traditional iris segmentation method based on contour matching and polynomial fitting,and an improved iris image segmentation method based on improved U-Net.In the iris segmentation method based on contour matching and polynomial fitting,the iris image is first divided into ideal and non-ideal types;then the circular shape of the iris is detected by using a circular filter on both the ideal iris image and the non-ideal iris image.The boundary is used to complete the ideal iris segmentation.For the non-ideal iris image,after detecting the circular iris boundary with occlusion,the polynomial fitting method is used to remove the occluded part,and finally the non-ideal iris is obtained.The accuracy of iris image segmentation by this method is 93.65%.In the improved iris image segmentation method based on improved U-Net,the dilated convolution is added to improve its performance based on the original U-Net network.Firstly,four feasible schemes are proposed,which are PD-Unet1,PD-Unet2,PD-Unet3 and FD-Unet.Then the four feasibility schemes are trained and tested on the same iris dataset,and the FD-Unet network with the optimal performance and the best segmentation is obtained.Thenetwork and the original U-Net then are trained and tested on the near-infrared datasets:CASIA-4i,ND-IRIS-0405 and visible light datasets: UBIRIS v2 to further verify the better performance of the FD-Unet network.The iris image segmentation accuracy of the proposed FD-Unet network reaches 97.36%.After the iris image was performed the segmentation processing,it is used to perform iris-image-based gender prediction.This paper proposes a method for gender prediction of iris images based on texture features and deep learning.Firstly,the texture feature of iris is extracted by using local binary pattern in segmented iris image.Secondly,a neural network with a two-level structure is constructed.The first-level network consists 8 layers,and the dilated convolution and no-padding convolution are used alternately to extract more iris features.The output of the first-level network and the iris texture feature map extracted by local binary pattern are used as the input of the second-level network.The second-level network is ResNet-34.The network is learned and trained,and finally,the result of gender prediction is reached.The method proposed in this article predicts men and women with an accuracy of 96.5%.Comparing with the iris segmentation algorithm and the iris image gender prediction algorithm proposed by other literatures,the methods proposed in this paper have better performance and can provide useful help for iris recognition.