Research On Face Spoofing Detection Technology

With the rapid development of computer vision and pattern recognition technology,facial recognition technology has been widely applied.Spoofing detection based on facial recognition technology has become a more concerned problem for researchers and users.Face spoofing detection is a technology to detect whether the face captured by the current camera is a living face,and it is an important guarantee for the security of the face recognition system.Especially in this age of multimedia,personal photos and videos are extremely easy to be used by outlaws.Face spoofing detection can effectively prevent lawbreakers from deceiving the face recognition system,and then protect personal privacy and property security.This thesis mainly conducted research on face spoofing detection,and had made the following improvements in the generalization ability of people’s face spoofing detection.1.A face spoofing detection method based on feature space constraints was studied.Through the difference of feature distribution between living face feature space and spoofing face feature space,living and spoofing were distinguished,and Euclidean distance was used to constrain the distance between the living face and the spoofing face,which minimized the distance between data of the same category and maximized the distance between data of different categories.To a certain extent,the problem of small inter-class differences and large intra-class differences was alleviated.In addition,from the feature point of view,this constraint method could promote the classification characteristics of low-intensity and weak classification features,and improve the generalization ability of the model to a certain extent.In the thesis,several groups of experiments were designed to verify the effectiveness of the method,and good classification results were achieved on multiple open data sets.Especially in cross-dataset experiments,the improvement effect of Euclidean distance constraint was more obvious.2.A hybrid-domain attention method was proposed to improve the generalization ability of the model.Hybrid-domain attention combined different focus domains,so that the training model could learn more feature information and focus information under the environment of the domain during training,so as to better cope with the complex real environment.In this thesis,we combined spatial domain attention and channel domain attention,and used the mixture of the two domains to verify the method on multiple data sets.The experimental results showed that the face spoofing detection method based on mixed domain attention had a good improvement effect.3.In this thesis,a face recognition sign-in system with spoofing detection was designed and implemented.The system mainly included three modules: “face registration”,“face check-in” and “log and information management”.The face registration module was to store the registration information and face image to the registration database;The face check-in module matched the captured face image with the registered face in the database to verify the user’s legitimacy;The log and information management module was to view and delete user information as well as records.There was a face spoofing detection in both face registration module and face check-in module.It made a spoofing judgment on the captured face firstly,and the system was only allowed to register and check in after judging that it was a living face,otherwise,the operation of registration and check-in could not be carried out.