Research On 3D Residual Local Binary CNNs Based Face Liveness Detection

With the wide application of biometric verification technology,the corresponding illegal intrusion also poses more and more serious threat to social security.Therefore,it is very important to study the face liveness detection.Many biometric detection algorithms,whether based on deep learning or traditional method,either because of insufficient data sets or inadequate extraction of feature information,are unable to fully learn spoofing/living information with discriminating and generalized capabilities,and resulting in poor practicability,finally.This paper proposes face liveness detection algorithms based on the influence of both time and space.Firstly,according to the classical local binary mode method,the paper gives a face liveness detection algorithm based on traditional dynamic texture feature VLBP operator,which is less computation and easy training;Then based on the deep learning,a face liveness detection algorithm based on 3D-ResCNN(Three-dimensional residual convolutional neural network)is improved;Finally,Combining VLBP with 3D-CNN(three-dimensional convolutional neural network),the algorithm of 3D-ResLBCNN(three-dimensional residual local binary convolutional neural network)for face liveness detection is proposed.3D-ResLBCNN is a combination of 3D-CNN and LBCNN(local binary convolutional neural network),adding residual idea.It not only takes into account 3D feature information including time dimension,but also combines local binary patterns to advance face information through LBCNN,which makes up for the deficiency caused by insufficient training data.In order to make the algorithm more generalized,the data are enhanced in different ways.Experiments on related data sets show that the proposed algorithm has obvious advantages in computing power and accuracy.