Research On Face Liveness Detection Method For Anti-video Spoofing

The current face recognition system can distinguish different user face images,but it is impossible to judge whether the face is derived from a real face or a deceived face in a photo or video,etc.,and is easily deceived.In order to enhance the security of the face recognition system,the face liveness detection has become a new research topic,aim to determine whether the face detected by the recognition system is from a living body,and then consider whether to open the face recognition system.For common face video attack methods and scenes,starting from the motion consistency of the background area and the texture information of the human face area,a new method of face liveness detection based on anti-video spoofing is proposed.The main work has the following three aspects:Aiming at the global jitter phenomenon in the handheld spoofing video,a face non-living detection method is designed based on the motion consistency of the background area.The new method divides the background contrast area according to the three directions outside the face detection area,and uses the optical flow statistical feature of the background area to detect the motion similarity of the contrast area to detect the face video attack played by the handheld device with the jitter phenomenon.Focusing on the image quality changes during video attack and the reflection of the display device,starting from the face area,the face liveness detection method with edge and local texture is designed for the edge detail blur and local highlight in the video.extracting the histogram of oriented gradient and the local binary pattern in the face image,and uses the joint feature of the two as a significant variation feature of the secondary imaging,and trains the SVM classifier to detect the real face and the video deceiving face.The new method was tested by Replay-Attack database.The results show that the combination of the two methods can effectively identify the common video attacks and realize the face liveness detection.