A face detection method based on multiple models

Abstract

A kind of human face liveness detection method based on multiple models, its steps are as follows: (1) select training sample and mark; (2) human face liveness detection model training based on multiple models; (3) human face liveness detection model conversion; 4 ) to carry out anti-photo attack human face liveness detection, the detection passes and enters step (5), otherwise it is considered that there is no living human face in the image, and the whole process is ended; (5) anti-screen attack human face liveness detection; if the detection passes, there is a living human face , otherwise there is no live face. The present invention uses three live face detection models and one face positioning model to detect live face, so that it can not only defend against photo attacks, but also defend against screen attacks, and the ability to defend against attacks is enhanced. According to the characteristics of the screen face image collected by the infrared camera, the face positioning model is used to replace the face liveness detection model for face liveness detection, which further improves the accuracy of face liveness detection.

Description

technical field [0001] The invention belongs to the technical field of face recognition, and relates to a multi-model-based human face living body detection method. Background technique [0002] Face recognition has been widely used in the fields of security and finance, and face liveness detection is the prerequisite for the application of face recognition. The function of live face detection is to detect whether there is a live face in an image or video, and to defend against attacks on face recognition-related applications from photos, screens, videos, masks or headgear. Existing face liveness detection methods include interaction-based liveness detection, stereo-based liveness detection, subsurface-based liveness detection, and deep learning-based liveness detection. Among them, the liveness detection method based on deep learning is the most widely used because of its robustness and detection accuracy. Among them, the interactive liveness detection requires the cooper...

AI Technical Summary

Problems solved by technology

[0004] 1) A single camera is used to collect face images and train a living body detection model, and the accuracy of living body detection needs to be improved;

[0005] 2) Existing liveness detection methods can only prevent photo attacks, not screen attacks, and the means of defense are single

Images