Face Detection Method Based On Extremely Short Video

Face recognition technology is one of research hotspot in the field of pattern recognition and has been widely applied in various fields.To address the various issues in the existing live detection methods,we propose a face live detection method based on very short video.First,Euler video magnification algorithm is combined with SIFT feature matching algorithm,as well as DCT transform to extract the features of static very short face video.Then,for the dynamic face video,we propose the Gabor-based LTP operator to extract the features.Finally,we use SVM classifier to complete the live detection.Therefore,this paper focuses on the face live detection from very short video.The main work is as follows:1.In view of whether the input face video image is slightly swaying or static,we propose to match the feature points between the frames in the video with SIFT feature matching algorithm based on RANSAC.Furthermore,we judge whether the video is static or not with the average distance between feature points.2.Aiming at the problem of face image detection in static state,Euler video magnification algorithm and SIFT feature matching method are proposed.First,Euler magnification technology is used to filter and magnify the static very short video.Then,SIFT method is used to extract the feature matching points of each frame of the original very short video.Finally,the red component of the enlarged Euler image is extracted.The coordinates of the extracted feature points are corresponded to the red component image,and the red points of all the feature points of each two frames are calculated in turn.The difference of quantum pixels is used to construct feature histogram,and SVM classifier is used to discriminate the real and false faces in static state.3.Aiming at the problem of false discrimination in static face live detection,a DCT transform algorithm based on high and low frequency coefficients of face is proposed.Because there is a small change in blood flow in real face,but there is no such change in false face,we extract the red component of real face and false face image to count the gray level pixel histogram of the image as the high and low frequency coefficients of face.Then,we use DCT transform to process the high and low frequency coefficients histogram to get the high and low frequency energy value after transformation,and finally,threshold is set as discriminant feature to solve the problem of false matching.4.To solve the problem of dynamic face live detection,a LTP algorithm based on Gabor filter is proposed.Firstly,the improved LTP algorithm of LBP operator is used to partition and describe the three-valued texture of face image to enhance the difference of texture features between real and false faces under slight shaking.Then,the texture features are filtered by Gabor filter in space and frequency to further enhance the contrast of the true and false face images and generates the feature histogram.Finally,the SVM classifier is used to detect the slight shaking in video.