With the advent of the era of big data, personal information security problems have become increasingly severe, based on face recognition technology to detect and image processing has been widely used. However, face detection technology are targeted at a small number of face images, with the deepening of the concept of big data, image data processing large face recognition technology will make higher demands. In the most primitive based face recognition system based on the current ratio between the person's face photo shoot with people face pre-stored photos, to authenticate. However, when the photos are fake himself when placed in front of this camera authentication systems face photo alignment based on this authentication systems based on facial photos may be verified by comparing the user's identity. In other words, a malicious user can use the photo counterfeiters to carry out malicious attacks (ie, photos attacks), which based on facial recognition system than the photos on the photo can not resist the attack. So, face detection technology came into the living body.
Attacks on face recognition system, there are three main categories: Photo attack, attack videos and 3D models attacks. Molecular illegal or counterfeit legitimate user after obtaining the user's photos or video, using a legitimate user's photos or video as a fake face trying to cheat the system. In order to distinguish the true face as well as photos, video, face recognition system to prevent possible attacks on, we need to apply face detection technology in vivo.
Face detection in vivo is mainly carried out by identifying the physiological information on the living body, the physiological information as it is vital to distinguish between a photograph biometric, silica, plastic and other non-living matter forgery. To make sure you are "living you" in vivo detection face discrimination step typically comprises several, for example blink determination: For applications may require the user to fit, requires the user to blink once or twice a face recognition system will automatically determine Zhang engagement state changes obtained from the eye to distinguish between pictures and face; pinch portion or the mouth determines: similar to blink determination, requires the user to open, closing a mouth twice a face recognition system and accordingly distinguish photos real people face.
Face detecting living body mainly includes: face detection, 3D detection, in vivo detection algorithm, continuity testing. Here explain separately.
Face Detection: Face positioned where, if there is no face detection process in vivo situation than the face, which can effectively prevent or switchover two people in the photos.
3D Detection: Verify whether the acquired three-dimensional portrait, picture plane can be prevented, the degree of curvature of the different photos.
Vivo detection algorithm: determines whether the user operation is normally done by a random operation specified user (shaking his head, nods, gaze, blinking, moving up and down the phone), a video prevent attack, abnormal operation of the attack.
Continuity Detection: in vivo and face detection can be used simultaneously better way to prevent people switching. Verify whether a face normal trajectory, if substitution occurs halfway abnormal movement; from the perspective of security can be prevented to skip directly detected face vivo replacement photo collection.
Several face discrimination step generally comprises detecting the living body, such as:
Blink determination: For applications may require the user to fit, requires the user to blink once or twice, in vivo face detection system to differentiate between photos and the face sheets according to the changes of engagement state obtained automatically determine the eye;
Determining pinch mouth: similar to blink determination, it requires the user to open, closing a mouth twice, face detection system in vivo and accordingly distinguish real facial photograph.
There are more such as shaking his head, nodded, looked up instructions and other movements, in order to identify and photograph real people face.
Currently, most face recognition schemes are based on direct extraction of facial image information, non-interactive, anti-attack ability, such as photos, videos, models camouflage can crack, this time, the living face detection technology the importance of self-evident.