Iris Liveness Detection With Convolutional Neural Network

With the advent of the information age,the rapid development of artificial intelligence,biometric identification technology plays an increasingly important role.Iris recognition technology is one of the most important technologies in biometrics.The human iris is a remarkable biological feature with rich texture features and high uniqueness and stability.Despite the rapid development of iris recognition technology,the iris recognition system will also be attacked by false iris.Iris detection is an indispensable module in iris recognition system.It can reduce the risk of iris recognition system being attacked by false iris.Convolution neural network is an important learning model of deep learning.Currently,it has achieved good results in face recognition and fingerprint recognition.In this article,convolution neural network is applied to the iris liveness detection task,and focused on the convolution of traditional neural network model on the basis of improvement.Convolution neural network was applied to the iris image feature extraction and classification.The experiment prove that the improved convolution neural network model has better performance in the iris live detection.The main contents of this paper are as follows:(1)We propose a convolution neural network iris liveness detection algorithm based on batch normalization.The algorithm utilizes the iris segmentation and normalization to preprocess the iris image,and then extracts the iris features via batch normalization convolution neural network(BNCNN).The decision-making layer distinguishes accurately the real iris and pseudo iris.Batch normalization convolution neural network can solve the ploblem that traditional approach of the iris feature extraction is usually too single,which leading to low identification rate.And it can solve the problem which is that the existing convolution neural network of overfitting and gradient disappearance.The experimental results show that the proposed method can extract the iris deeper texture feature,and achieve higher accuracy rate compared to some typical iris liveness detection methods.(2)This paper proposes a new YOLO based iris liveness forensics technology.To accur ately index the true and false images,in YOLO,the convolution layer is used to extract the candidate box of the iuput imagese,while the global average pooling is employed for pre-clas sification.The YOLO algorithm can subtly solve the high computational complexity proble m faced by convolution neural network as well as accurately identify the true iris and fake irises.Experimental results demonstrate that the YOLO algorithm not only consumes less computational time but also achieves higher recognition performance,compared to the convol ution neural network.