Face Liveness Detection Based On Texture And Deep Learning

With the advancement of technology,more and more mobile devices are equipped with front cameras.At the same time,due to the easy collection of facial features and the ease of use,facial features are more and more widely used in people's daily lives,including facial unlocking,Pay by face,etc.However,because the face pictures are easy to obtain,and even downloading pictures from the Internet can be used to spoof the face recognition system,it is necessary to enhance the security of the face recognition system and increase the liveness detection function of the face.This thesis focuses on face liveness detection based on texture features and deep learning.The main contents are as follows:(1)In order to demonstrate the necessity of face live detection,a questionnaire on face unlocking and face payment was designed.From the perspective of data analysis,a descriptive statistical analysis of the subject's risk measurement was conducted,and the side demonstrated the necessity of face liveness detection.(2)This paper proposes a face liveness detection method based on texture features.First read the image and convert it into a grayscale image,adjust the image size to 128x128,and then use local binary mode(LBP),Haralick texture features,RGB color histogram to extract image texture features,and use principal component analysis to local binary mode,the extracted features are processed for dimensionality reduction,and finally feature fusion is performed,which is fed into a classifier to build a model.The ROC curve shows that the proposed algorithm has excellent classification performance in NUAA and Replay-Attack face cheating databases.(3)This paper proposes a model combining traditional texture feature local binary pattern(LBP),histogram of gradient(HOG),and convolutional neural network,and feeds 0.6 * LBP + 0.4 * HOG features into the convolutional neural network for learning,experiments show that the classification accuracy of the face cheat database in NUAA is high.