Research On Face Liveness Detection In Intelligent Access Control System

Face recognition authentication technology for its convenient,non-contact and other characteristics,is widely used in security,access control,attendance and other fields.However,the traditional face recognition system can recognize different human faces,but it is difficult to determine whether the face for living or photos,which will bring great safety hidden trouble to identity authentication system.Therefore,the human face living detection has been widely concerned and become a hot topic of research.For photo cheating problems,two methods to resist photo cheating,no user cooperation,no additional equipment are proposed combining with machine learning theory,based on image texture information by summarizing the domestic and international research of face liveness.The main work of the paper includes the following two aspects:Focusing on the nonlinear variations of photo face after secondary imaging which formed the difference of surface texture information.Starting from the details change of image edge detection,we designed a face living discriminant method.The new method using gradient direction histogram to count texture information and respectively making the average gradient direction feature of the real faces and the photo face as a primary reference,and then combined with the histogram intersection to compare the similarity between the object to be identified and the reference feature to construct the global quantitative description.Finally,we successfully used the SVM classifier to classify real face and photo face.Aiming at the problem that the artificial selection feature is difficult and theabstract features are easily neglected in the face living discrimination,we have designed a LBP features combined with the depth of learning of face living discriminant method.This method uses the depth belief network as the main model and takes the LBP feature as the input of the depth network to highlight the real face and the photo face in the local texture difference,that is,from the bottom to up to automatically learning the real face and photo face with a distinguishing abstract features,but also can improve the problem ignoring the two-dimensional local structural feature when the image is entered into the depth network with pixel characteristics.