Face Spoofing Detection Algorithm Research Based On Texture Analysis

Face Authentication System(FAS)are vulnerable to spoof attacks made by fake faces.Thus,face spoofing detection become a vital part of FAS.Texture feature is reflected on the basic information of face images,faces are took from the cameras,and texture feature mainly identify real face and fake face in the anti-spoofing detection.However,researchers on non-intrusive software-based face spoofing detection have mainly been focusing on analyzing gray-scale images.Hence the color texture information has been discarded,which can be a vital visual cue for discriminating fake faces from genuine ones.Main works of this thesis include as follows:Firstly,we propose a robust and appealing local descriptor,called WLDLBP(Weber Local Descriptor and Local Binary Pattern).First,we analyze the classification ability of texture feature in the multi-color space,and verify that the performance of multi-color feature space fusion is better than the single color space.Then each histogram of WLDLBP and LPQ was extracted respectively and concatenated orderly to a new row histogram by exploiting the joint color texture information from HSV,YCBCR and RGB color space.Finally,a library for Support Vector Machines classifier(LIBSVM)is trained to distinguish the genuine and spoofing faces.Our proposed method carried out with CASIA Face Anti-Spoofing Database and Replay-Attack Database show that experimental results attained more stable and more robust performance than the state-of-the-art techniques in both intra-databases and inter-databases,and can be widely applied to the non-intrusive software-based face spoofing detection.Secondly,we propose a dynamic local texture feature algorithm(called WLDLBP-TOP)from three orthogonal planes in multi-color space.Dynamic Texture(DT)is an important research method to face spoofing detection.The feature of three orthogonal plane and multi-color space are merged,and then the characteristics of three orthogonal plane in multi-color space are researched.The results show that the proposed algorithm is more effective than the state-of-the-art techniques based on dynamic texture.Last but not least,we designed a collected face and liveness detection system.The system contains registration stage,face collection stage and face anti-spoofing stage.Face collection stage contains face detection and collection.The face anti-spoofing stage contains face detection,feature extraction and SVM prediction.