Research And Design Of LBP And Challenge Response Based Liveness Detection For Face Authentication

With the continuous development of technologies such as machine learning and big data,the face recognition system based on identity authentication is more and more widely used,because it has many advantages,such as memory-less authentication process.However,it has an intrinsic vulnerability against the media-based facial forgery(MFF)attacks where adversaries use photos and videos containing victims' faces to circumvent face authentication systems.In today's social networks,a large number of private photos and videos containing faces are uploaded every day.Adversaries can easily obtain these photos and videos to attack the face recognition system.The liveness detection module is an important part of the face recognition system,and it is also an effective defense technology to prevent the media-based facial forgery attacks.It can differentiate real live faces and fake forged faces.In this article,I propose a practical and effective liveness detection mechanism that combines a challenge-response protocol and local binary pattern operator to protect face recognition systems from the media-based facial forgery attacks.Our liveness detection algorithm consists of four modules,which are challenge-response module,expression frame detection module,feature extraction module and liveness classifier module.The challenge is first randomly sent to the user through a random number method,and then the response is recorded as a video using a camera.Challenges refer to facial expressions such as blinking,smiling.Local binary pattern operator is a method to describe local texture features.It is widely used in texture analysis due to its high efficiency.We use d-LBP(a variant of the local binary pattern)to detect expression frames in the response video.The expression frame refers to the frame with the most prominent expression characteristics.For example,in a blinking video,the expression frame refers to the frame where the eyes are completely closed.Next,we use the image gradient method to analyze the local gray value distribution in different subareas.We use a smaller sampling radius in the subarea with a larger image gradient and a larger sampling radius in the subarea with a smaller image gradient,then draw a local binary pattern histogram and integrate it into a feature histogram.Finally,the liveness classifier module detects the media-based face forgery attack by comparing the similarity called(35)between the input histogram and the real texture feature histogram.If(35)is greater than the threshold,it means that the input face comes from a forged face of the media-based facial forgery attack.In order to prove the effectiveness of this liveness detection algorithm,we collected real face data set from legitimate identity authentication requests,and also collected false face data set based on the media-based facial forgery attacks.The fake face data set includes a photo-based data set and a video-based data set.The final experimental results show that this liveness detection method can effectively detect the media-based facial forgery attacks with an accuracy of 96.53%.Since this liveness detection method obtains the texture feature information of the human face,it can also be used to recognize the faces of different users,and the recognition accuracy rate reaches 98.96%.