Research On Face Live Detection Algorithm Based On Deep Learning

With the continuous progress and improvement of science and technology,intelligent recognition systems have penetrated into people's lives and work,including: intelligent face recognition systems,intelligent fingerprint recognition systems,intelligent iris recognition systems,and so on.Identity authentication is becoming more and more important,and it is accompanied by various fraudulent methods of falsifying legitimate user information.The rapid development of digital and intelligent identity authentication and recognition has gradually attracted attention to security issues related to identity authentication.The current face recognition system cannot detect human faces,and it is impossible to determine whether the face in front of the camera is a real person face.Face fraud methods mainly include face video fraud,face photo fraud,threedimensional simulation face fraud and other methods.What needs to be done for live detection is to prevent the above-mentioned fraud methods from cracking the face recognition system.Provides face detection function for face recognition system,making authentication of identity information more secure.The paper summarizes the current state of the art research on face live detection,addresses the problem of fraud methods in videos and pictures,and conducts research from the facial feature information in the images.According to the methods of picture and video fraud,combined with deep learning,it designs photo fraud prevention There are two methods of video anti-fraud.This method does not require user cooperation and does not need to be equipped with face live detection methods.The main work of this paper is from the following two aspects:(1)Perform feature analysis on photos of faces.Photo fraud is mainly through secondary imaging of face photos.Living faces are a complex three-dimensional structure.The reflected light intensity at each angle is different,which will cause different Reflections and shadows,a person's face photo is a flat structure,and a secondary imaged face is generated from a person's face photo,so facial features are significantly different.The features of real face pictures and false face pictures are extracted by Convolutional Neural Network(CNN),trained and verified,and finally real and false face classification is realized.(2)Feature analysis is performed on the video of the face.The video fraud method mainly uses video to simulate the real face,and the fraud video can be the second shot of the face picture or face video.The living face is a three-dimensional structure,and the false face video is a two-dimensional structure after secondary imaging.The dense optical flow method is used to calculate the optical flow information in every two frames of the video.The motion information of the human face light can be calculated.Each frame of the video calculated by the optical flow method is saved,and feature extraction is performed by a convolutional neural network(CNN),and finally the real face video and the fake face video are distinguished.