Research On Key Technologies Of 2D-3D Face Detection Based On Texture Feature

Face recognition is a biometric technology with high precision,good stability and convenient used.However,face recognition technology is frequently subjected to fake attacks(or replication attacks),there are still many security risks.Live detection which has a significant effect in terms of resistance to counterfeiting attacks(or copy attacks)identified whether the sample has a life feature.Aiming at the problem that the face recognition system cannot recognize whether the collected face image is from the real person,this paper focuses on the face living detection algorithm based on 2D face image and 3D face depth map.The main tasks include:1.In view of the problem that the existing 3D face detection database is less,this paper collects a RGBD face database.The positive sample of this database including the 104 real face of 20973 pictures at different depths by Kinect and another binocular device.The negative samples which collected iPad,computers,mobile phones and photos attack face by Kinect comprised 12300 pictures in different environments at different angles of 0.5-2 meters.2.In this paper,an improved Fourier spectrum feature method is proposed to solve the problem that the existing Fourier spectrum analysis method is relatively simple and the accuracy is low.In this method,the Fourier spectrum is divided into several sub-blocks by the method of block subspace,and the average energy value of the image in each sub-block is obtained,normalized and cascaded into a global Fourier spectrum feature Vector.Experiments show that the improved Fourier spectrum feature can effectively improve the accuracy of 2D face anti-spoofing.3.Aiming at the increase of training samples,the accuracy of 2D face detection based on Fourier spectrum features will be further reduced.This paper proposes a FS-LBP feature face detection method based on LBP features.This method combines the Fourier spectrum characteristics and low-dimensional LBP features and used SVM to classify and judge the results.Experiments show that the method is better than the mainstream method of MSLBP feathers.4.Aiming at the low level of GLCM and the detection rate of 3D faces can be further improved,this paper proposes a method of multi-scale GLAM.Firstly,the face image of the depth image is adjusted to the depth image with different scale size by face detection of the RGB image and the face region image of the depth map is acquired synchronously.Then,the GLCM feature is extracted and Combined with a multi-scale GLCM features,and finally use SVM to classify and judge the results.Experiments show that the accuracy of the method is higher than that of the GLCM and the LBP feature in the 3D face anti-spoofing.Finally,the work of this paper is summarized,and the follow-up work of this paper is prospected.