The Research Of Face Liveness Detection In Identity Authentication

In recent years,biometric authentication technology has been developing rapidly.Among them,face recognition is widely used in authentication of various security systems.However,the traditional face recognition system can only identify the face in the collected image,but it can not distinguish whether the face is from the real face or a print out of the photo,or pre-recorded video in front of the camera.In general,the traditional face recognition system have drawbacks which not be ignored.It can not distinguish between the real face and fake face,that is,can not distinguish the face whether is real or fake.In order to solve this problem,face anti-spoofing detection technology came into being.However,due to the variety of face attacks mode and different environment in face recognition,face anti-spoofing detection technology has become a difficult and hot spot in the field of face recognition.The purpose of this thesis is to study face anti-spoofing detection technology,and we study the research of predecessors in related fields.The main work of this thesis is as follows:(1)Face Anti-spoofing Detection Method Based on Kinect Depth Information:In view of kinect can effectively and inexpensively to obtain the characteristics of the depth information of the image,this method use kinect to capture color image and depth image,then extract texture descriptor,such as the local binary mode operator,and combine with binary classifier to identify the face whether is real or fake.In addition,to evaluate the face anti-spoofing detection performance of this texture descriptors,we compare different types of LBP operators and their variants.Considering that the smaller area of the face may may make the texture features of the image attack or video attack more visual,we propose to use two different methods(face segmentation and face segmentation)to calculate texture features.And the performance of true and false face classification of different binarization classifiers is compared.Considering that the face anti-spoofing database has been published can not meet the experiment requirements,we use kinect to collect the face depth information to form the new data set,and verify the performance of the algorithm in the dataset.(2)Face Anti-spoofing Detection Method Based on Convolution Neural Network:This method uses the complex and effective convolutional neural network(CNN)to study the features of real faces and fake faces,and then combines with the support vector machine to distinguish the face whether is real or fake.This approach does not need the user interaction and does not require additional hardware equipment.In addition to using CNN to learn effective feature,we also propose two data enhancement strategies,including the increase of spatial scale and the increase of time scale.By increasing the spatial scale,we further explore the effect of background region on face anti-spoofing detection,and use the fusion of spatial scale and time scale to improve the performance of face anti-spoofing detection.