Research Of Some Key Technologies On Face Occlusion And Spoofing Detection

Intelligent video surveillance system and face detection application are now used more and more widely.New technology has brought us many convenient and liberated human labor powers,but also brought new security issues.1)Face occlusion attackWith the rise of crimes associated with ATM,face occlusion detection has gained more and more attention because it facilitates the surveillance system of ATM to enhance the safety by pinpointing disguised among customers and giving alarms when suspicious customer is found.However the existing face occlusion detection algorithms have poor robustness in the complex environment,such as the variety types of occlusion,and the large occlusion area.In that condition,the existing deep learning model is inefficient and can not realize real-time detection.2)Fake face spoofingWith the popularization of the face recognition system application in payment,personal account login,family entrance guard and other high level security requirement environments,criminals may use printed photos to imitate users' faces so as to attack the face recognition system which will result in unexpected loss.So it is necessary to add face liveness detection in face recognition system.However the performance of existing face liveness detection algorithms are easy affected by the illumination,the image resolution and the fake face material,unable to meet the requirement of practical application.Although deep learning has made great success in the field of computer vision,it is rarely applied to face occlusion detection and face liveness detection.In order to improve the security of face recognition,this paper proposes a novel algorithm based on deep learning:1)Proposes a cascaded Convolutional Neural Network(CNN)based face occlusion detection method,which enhanced the robustness of algorithm under complex conditions by strong learning ability from data and high efficient feature representation ability of deep learning.It also improved the algorithm processing speed by the cascade structure.2)Proposes a face liveness detection algorithm based on Multi-task Convolution Neural Network(MTCNN),which is possible to simultaneously detect the state of the eyes and the mouth.The structure,which is combined with depth and shallow information,can retaining the details of abstract features,but also considers the global shape features.As a result,the precision and robustness are improved.Furthermore,facial pose estimation using the PnP algorithm combined with key points tracking algorithm can effectively improve the accuracy.The face liveness detection system adopt the interactive model to judge the face liveness by detecting the completion of the corresponding instructions,such as blinking eyes,turning head and opening mouth.Experiments carried out on several public databases and in-house databases indicate that the proposed algorithm has strong adaptability in the complex environment,and has high recognition rate for diversity type of face occlusion and fake face spoofing.