Research Of Face Liveness Detection Algorithm Based On Face Recognition System

With the continuous development of modern science and technology,the face recognition system has gradually replaced the traditional security identity authentication system and was widely used in various security fields.Liveness detection technology plays a vital role in face recognition system.In the modern society,images and videos containing legitimate user face information are easily stolen,and the technology can distinguish the authenticity of the current user and improve the overall security of the system.Currently,the face liveness detection technology could be divided into user collaborative method,relying on special equipment method and silent detection method.There are some obvious defects in the first two methods,such as poor user experience,long detection time,high equipment cost,and limited application scenarios.The silent detection method is the most ideal method.For the research focus of this kind of algorithm,this paper has carried out sufficient research work,mainly as follows:(1)In view of the problem that the texture-based algorithms ignore the color feature information,this paper proposes to replace the traditional gray space and RGB color space with the fusion color space.In this paper,the color channel correlation experiment and the contrast of the color distribution of the true and fake face images in the fusion color space are used to verify the validity of the color feature information for the face liveness detection algorithm.(2)Aiming at the problem that image classification features in previous algorithms are not efficient enough,this paper proposes a fusion color texture feature.This feature adopts two features of Co-Occurrence of adjacent Local Binary Pattern(CoALBP)and Local Derivative Pattern(LDP)to realize the fusion feature and feature extraction in the fusion color space.Therefore,the fusion color texture feature includes color feature information,texture feature information,spatial feature information,and gradient feature information.(3)For the problem that the previous algorithms perform well in the internal verification of the database and perform poorly in the database cross-validation,this paper proposes a method of extracting the feature by using high-discriminative local image patches instead of global image.In this paper,five correlation algorithms are proposed to select high-discriminative local image patches from candidate image blocks,namely DEND cluster,CP(Cluster pairing)algorithm,CS(Cluster space)algorithm,MAXDIS algorithm and IQA(Image Quality Assessment)algorithm.And related experiments were carried out,and the most suitable algorithm was selected by analyzing the experimental results.In order to verify the effectiveness of the proposed algorithm,relevant experiments were designed and carried out.For compare with the past excellent algorithms,four standard public databases will be used,and they are Replay-Attack,CASIA,MSU and OULU.The experimental results show that the proposed method has achieved great results in both internal verification and cross-validation experiments.