Research On Face Spoofing Technology In Face Recognition System

With human beings entering the information age,face recognition solutions have been widely used in various fields,and its security has attracted more and more attention.Because of the popularity of social networks,it is easy to obtain human face images to attack the face recognition system.So in the past few years,human face detection technology has become one of the most popular research directions in the field of human face,which has a strong practical value.In this context,the main work of this paper is as follows:In view of the loss of texture information in the process of face secondary acquisition,this paper proposes a method of combining hand-designed features and depth features extracted based on convolutional neural network.In the past,hand-designed features are mostly used in human face detection.However,hand-designed feature extraction methods are based on prior knowledge and have certain limitations,which leads to detection precision However,the convolution neural network is more powerful in depth feature representation,which can be directly learned from input image observation by data-driven way.In order to overcome the limitations of hand-made features,this paper proposes to combine hand-designed MLTP features and depth features to form a hybrid feature.The hybrid feature contains more sufficient texture information and has stronger discrimination ability.Then,PCA method is used to reduce dimension and remove redundant information,and SVM classifier is used for classification.The algorithm proposed in this paper can make full use of the advantages of MLTP features and depth features,improve the detection accuracy and increase the robustness of the algorithm.In order to increase the stability of human face detection algorithm,this paper proposes a human face detection algorithm based on static texture feature and motion feature,and designs a multi input depth separable convolutional neural network framework for human face detection In order to extract the dynamic information of human face,the face images in RGB color space and HSV color space are used to extract the static texture information of human face.The final classification results are obtained by weighted combination of three subnetworks.The algorithm proposed in this paper has better robustness when dealing with different attack methods.The method of decision fusion with multiple complementary clues,such as static features and dynamic features,can make full use of the advantages of various features to achieve better performance and is more suitable for the actual complex scene.Finally,this paper evaluates the effectiveness of the proposed algorithm in NUAA face database,display-attack face database,CASIA face database and 3D mad face databaserespectively,and compares it with the existing algorithm.