Research On Liveness Detection For Face Authentication

In recent years,biometric authentication technology has been developing rapidly.Among them,the face recognition technology based authentication has been widely used in various practical scenarios.However,the face recognition technology is mainly focus on the identification of the face in the image but could not distinguish whether the face image is from real face of valid user or from fake face forged by illegal users.Nowadays,with the advanced Internet technology,it seems easy to get a face image of a valid user.Therefore,the security of the face authentication system is under an unprecedented threat.Once the illegal users successfully attack the face authentication system,the consequences and loss may be extremely serious.In order to solve this problem,face liveness detection technology came into being.The main purpose of face liveness detection is to identify whether the face image collected by the face authentication system is real or fake.Due to the diversity of forgery and fraud methods and the uncertainty of external environment,the technology of face liveness detection is confronted with great challenges.In this paper,some studies have been done based on the analysis and summary of the existing liveness detection.The main work and innovations are as follows:1.A face liveness detection algorithm based on illumination distribution is proposed.Affected by the surface properties of the display media and the ambient light during the recapture process,the fake face images tend to show different illumination effect from the real face image.Based on this observation,a face liveness detection algorithm based on illumination distribution is proposed.First,the face image is transformed into different color spaces and specific channels are selected and filtered for a certain number of iterations utilizing nonlinear diffusion model.The diffusion speed feature is obtained according to the diffusion image and the original image.Then the diffusion speed features of each channel are fused to generate the multiple color space diffusion speed feature of the image.To fully describe the illumination distribution of an image,the specular ratio is used to describe the characteristics of the highlight region.The specular reflection component of the face image is obtained by the highlight removal algorithm.The specular ratio feature is calculated according to the proportion of the specular reflection component.The illumination distribution is more comprehensively described by combining the diffusion speed feature and the spacular ratio feature.Experimental results on CASIA-FASD and REPLAY-ATTACK fully prove the effectiveness of the algorithm.2.A face liveness detection algorithm based on dynamic texture analysis is proposed.The fake face cannot reproduce the real face without distortion in the printing or display process.Moreover,the surface properties of the display media of fake face are quite different from real skin.Therefore,the fake face images collected by the camera have different micro-textures from the real face.According to this characteristic,a face liveness detection algorithm based on dynamic texture analysis is proposed.First,the existing texture feature descriptor LSP(Local Salient Pattern)is improved.Then,according to the idea of dynamic texture analysis,MLSP-TOP(Modified Local Salient Pattern from Three Orthogonal Planes),a new texture feature based on three orthogonal planes is proposed.It takes into account both spatial and temporal information with multiresolution strategy to extract dynamic texture features of image sequence.The experimental results show that the proposed method outperforms other state-of-the-art techniques based on texture features.