A Face Liveness Detection Method Based on Matv Feature

Abstract

The present invention relates to the human face detection method based on MATV feature, comprising steps: read and decompose the training set video, extract the human face area of ​​each frame and normalize as input, adopt the MATV light based on the amplitude-angle variation method The flow field features are encoded with statistical histograms to obtain the MATV histogram features, and finally the MATV histogram features are input into the SVM classifier for training, and the SVM model is established; for the face video frame sequence of the test set, each frame is Face detection, and then extract the face area and normalize it to get the face frame sample of the test set, calculate the MATV histogram feature of the face frame sample, and finally input it into the trained SVM model to predict the face liveness detection the result of. On the basis of the traditional optical flow method based on amplitude features, the present invention adds angle features and considers the directional characteristics of the optical flow field, thereby obtaining more movement details of living and non-living human faces, and improving the accuracy of human face detection Accuracy.

Description

Technical field [0001] The present invention relates to the research field of face live detection, in particular to a face live detection method based on MATV (Magnitude-Angle Total Variation) features. Background technique [0002] With the popularization of automatic face recognition technology, a series of potential security threats will follow. Criminals can attack and deceive the automatic face recognition system to gain benefits. There are mainly two common forms of attacks. One is to print a photo of the face of a legitimate user to attack, and the other is to use a playback medium to record a dynamic video of the face of a legitimate user to attack. If the criminals' attempts succeed, the consequences will be disastrous. [0003] In order to deal with this potential spoofing attack, face live detection technology came into being. Current live detection technologies are mainly divided into two categories: one is based on image texture features, such as LBP descriptor and ...

AI Technical Summary

Problems solved by technology

However, the traditional optical flow method is not good enough to estimate the optical flow of the motion information of living and non-living faces, so it cannot distinguish living and non-living faces well. At the same time, these methods often only consider the amplitude characteristics of optical flow and ignore Directional characteristics of optical flow field

Therefore, traditional optical flow methods do not achieve very effective liveness detection performance.

Images