Exploiting Temporal And Depth Information For Multi-frame Face Anti-spoofing

With the wide application of Internet technology and the in-depth development of computer vision,the reliability and efficiency of face recognition technology have been greatly improved and it plays an important role in the identity authentication system.Face recognition technology is indispensable in areas such as security,finance,cyber security,and attendance.At the same time,however,there are endless attacks on the face authentication system.Common attacks include taking photos of legitimate people or playing back legitimate videos of legitimate people,and with the rapid development of 3D printing technology,3D printing masks.The threat is beyond the traditional twodimensional attack,and the security of these face recognition systems poses a great challenge.By utilizing pixel-wise supervision,depth supervised face anti-spoofing reasonably contains more generalization than binary classification does.With-out considering the importance of sequential information in depth recovery,previous depth supervised methods only regard depth as an auxiliary supervision in the single frame.In this paper,we propose a depth supervised face anti-spoofing model in both spatial and temporal domains.The temporal information from multi-frames is exploited and incorporated to improve facial depth recovery,so that more robust and discriminative features can be extracted to classify the living and spoofing faces.Extensive experiments indicate that our approach can distinguish the real face from the image and video face effectively.With the development of 3D printing technology,the 3D mask spoofing attack has been becoming the new threat.On the basis of the shearlet transform,combining the geometric attributes based 3D facial description and local regional texture changes,a feature fusion method is proposed by using multilayer autoencoder network to identify the attack mask.The 3D face image to be registered are decomposed into a lowfrequency sub-band and several high-frequency sub-bands by the NSST method.We firstly construct the Gaussian scale space by smoothing the meshed face scan in the low-frequency sub-band and sign the location and direction allocation of feature points.Then,the features are combined in series and fed into the stacked autoencoder network,and the softmax classifier is employed to fine tune the network.The experiment on the BFFD database based on the flexible TPU material 3D print mask shows that the multifeature fusion method has greatly improved the accuracy of the anti-spoofing performance against mask attacks compared with the previous method of using the texture feature alone.Finally,a video level face in vivo detection system is designed and implemented to test and verify the effectiveness of the proposed algorithm and the follow-up work of this paper about new method is prospected.