Research Of Face Liveness Detection Based On Improved Convolutional Neural Network

With the upward trend in the maturity of computer vision and biometric authentication,face authentication technology has penetrated in all aspects of human daily life throughout latest years due to its uniqueness,relative stability,non-contact acquisition and other superiority.While this technology is widely used,in order to strengthen the safety elements of face detection and prohibited the system from being deceived and attacked by non-liveness faces,many researchers in this related areas pay much attention to the way of enhancing the system security by applying the face liveness detection.This paper for the face liveness detection is actually a binary classification issue,which is to analyze the essential difference between real face images and remake images.If the real face is not detected,the next program will not be executed,and this recognition will be regarded as an intentional attack.This paper first analyzes the essential difference between the real face images and the remake images.Then based on the deep learning framework,combines convolutional neural network to the face liveness detection.The core of this paper explores the following topics:(1)Conduct the preconditioning operation including image normalization and image geometric transformation towards each image sample in the face anti-spoofing sample database,so that the processed face image more closely matches the face liveness detection that needs to be studied.(2)Through detailed and meticulous study on current face liveness detection algorithm and convolutional neural network theory,this paper demonstrates the superiority of applying convolutional neural networks into the face liveness detection.The proposed convolutional neural network is specifically targeted at the drawbacks of conventional manual feature extraction,and it is quite applicable to the research problem proposed in this paper by improving the activation function and introducing batch normalization layers.(3)After the convolutional neural network proposed,then a heterogeneous kernel based on convolutional neural network algorithm proposed,which can not only ensure the accuracy but also effectively reduce the computational of the model.And at the same time,the optimized classifier can also make the classification effect better and more effective.(4)The algorithm in this paper was validated on the classic face anti-spoofing databases NUAA and CASIA-FASD.In the horizontal comparison and analysis of the experimental data before and after the improvement of the algorithm,and the longitudinal comparison of other face liveness detection algorithms,the paper finally proves that the proposed algorithm has better effect of reduces the amount of computation and the classification than the other algorithms.