A live face detection method based on wld-top

Abstract

The invention discloses a living human face detection method based on WLD-TOP, comprising the following steps: (1) training stage: reading training set video, performing human face area detection on each frame, and converting it into a gray scale human face Sequence of image frames, construct a three-dimensional image matrix, then construct a filter template and calculate WLD features, then generate WLD-TOP feature vectors, and finally input the feature vectors into the SVM classifier for training, thereby establishing the SVM model; (2) Test phase: For the test The image sequence of each frame is detected and converted into a grayscale face image sequence, then the three-dimensional image matrix and filter template are constructed, the WLD feature is calculated, and the WLD‑TOP feature vector is generated, and finally sent to the trained SVM model , to obtain the live face detection result. The present invention utilizes Weber's theorem, and on the basis of LBP-TOP, not only reflects the size relationship between neighboring pixels and central pixels, but also quantifies the difference between neighboring pixels and central pixels, making the features of the descriptor more comprehensive.

Description

technical field [0001] The invention relates to the research field of human face detection, in particular to a living human face detection method based on WLD-TOP. Background technique [0002] Face recognition technology compares and analyzes the biological characteristics of human faces to identify people's identities. Face recognition technology has made great progress in the past few decades, and face recognition products have been used in many occasions such as access control, monitoring of important places, and entry and exit. One of the advantages of face recognition technology is that it can automatically identify targets without supervision, but it also leaves hidden dangers. If criminals use users' photos or even videos to easily fool the face recognition system, it will cause harm and serious threats. to social security and stability. [0003] Common face spoofing attacks include photo attacks and video attacks. Photo attacks have the user's facial features, wh...

AI Technical Summary

Problems solved by technology

Photo attacks have the user's facial features, while video attacks have more dynamic features of legitimate users, such as blinking and changes in facial expressions, which are more deceptive and seriously affect the accuracy of the face recognition system.

[0004] The current living face detection methods mainly include the following types: one is based on the texture structure analysis method, which extracts relevant texture features for discrimination by analyzing the difference between the three-dimensional living face and the retaken face imaging; the other is based on In the method of facial motion information analysis, the essential difference between a living face and a retaken face is that the former is a three-dimensional object, while the latter is a two-dimensional plane structure. There is a second shot of the face, and the motion effects produced by them are completely different; The third is the method based on the analysis of living body feature information. This method analyzes the thermal infrared image of the face, eye blinking and lip movement and other living body features. This method may require some additional detection equipment support, so there are hardware limitations in promotion

Images