Multi-Feature Fusion Detection Method For Fake Face Attack In Identity Authentication

Face recognition has been playing an important role in the era of AI,with the promotion and application of brush face payment,the popularity of face card and the release of cell phone face.However,face recognition system is still faced with many threats because of the complexity of the face recognition environment and the continuous improvement of attack technology.Therefore,it is also necessary to study the face recognition technology in depth.Based on dynamic features and static characteristics,two kinds of multi-feature fusion algorithm,neighbor LBP and optical flow fusion,multi-channel linear LBP and optical flow fusion algorithm,are proposed,and DBN is used to learn and classify.Experiments show that all two algorithms have high recognition performance.The main research contents are as follows,1.Aiming at the difference of texture difference and dynamic characteristics between the two acquisition of real face and fake face,a fusion algorithm of LBP algorithm and optical flow algorithm based on neighborhood mode is proposed.Neighbor LBP modifies the traditional centering point contrast to clockwise neighbor comparison,and uses Gauss Pyramid model to extract features after image segmentation,which reduces the dimension of features and improves the accuracy of detection.2.A multi-channel linear LBP and optical flow fusion algorithm is proposed because of the loss of some color information on the gray level processing of the image.The RGB image is decomposed by three channels,and then arranged and combined.The three channel image is decomposed into eight different channels,and the linear LBP feature is extracted from the image.Then,the optical flow algorithm is applied to extract the feature.3.Because the SVM algorithm belongs to the shallow depth of network learning,belief network(DBN),limited Bohr Menzies machine(RBM).The feature extracted from the above multi feature fusion algorithm is used as input vector,and trained and optimized by DBN network,which is applied to the classification of real face and fake face.Tests on three kinds of data sets of CASIA,REPLAY-ATTACK and MSU are carried out.The experiment shows that the algorithm has good recognition performance.