When I saw a fun project on the weekend, I thought about actually using it. This project is mainly about open source content related to face liveness detection. The silent liveness detection method is mainly used here.
Face silent liveness detection is a technology used to verify that a human face is real and alive without any verbal instructions or specific actions. It prevents face spoofing attacks using photos, videos or masks etc.
The following is an overview of a common face silent liveness detection method:
-
Infrared illumination: This method uses infrared illumination to observe the human face. Real living human faces will produce weak infrared reflections due to blood circulation, while photos or masks do not have such reflections. Infrared sensors can be used to detect and differentiate these reflections.
-
Depth information detection: This method uses equipment such as a 3D depth camera or a structured light camera to obtain the depth information of the face. Real faces usually show more details and structures in depth information, while prosthetics or photos usually show relatively flat and single depth information.
-
Blood circulation detection: This method judges the vitality of the human face by observing the blood flow on the surface of the face. Real human faces have weak superficial movements due to the flow of blood across the skin, while prosthetics or photographs, etc. cannot produce this kind of movement.
-
Liveness detector: Similar to the methods of blink detection and mouth movement detection mentioned above, the liveness detector based on deep learning can also be used for face silent liveness detection. These detectors make judgments by learning the difference between real and fake faces.
It should be noted that using one detection method alone may not fully guarantee the accuracy and safety of liveness detection. Therefore, in practical applications, multiple different silent liveness detection technologies are often combined to improve the accuracy of judgment.
With the help of hardware and other equipment, the form cost is high and the environmental requirements are high. Many projects here are mostly based on the form of images to judge whether they are alive or not.
Liveness detection technology is mainly to determine whether the face appearing in front of the machine is real or forged. Among them, the face presented by other media can be defined as a false face, including printed paper photos, display screens of electronic products, and silicone masks. , Stereoscopic 3D portraits, etc. The current mainstream liveness solutions are divided into cooperative liveness detection and non-cooperative liveness detection (silent liveness detection). Cooperative liveness detection requires the user to complete the specified actions according to the prompts, and then perform the liveness verification, while the silent liveness test directly performs the liveness verification without the user feeling.
Because the Fourier spectrogram can reflect the difference between real and fake faces in the frequency domain to a certain extent, we have adopted a silent living body detection method based on the auxiliary supervision of the Fourier spectrogram. The model architecture consists of the classification main branch and the Fourier The spectrogram auxiliary supervision branch is composed, and the overall structure is shown in the following figure:
The official project address is here , as follows:
If you are interested, you can experience it by yourself. The open source caffe model is open here, and the speed is quite fast after experience.
Here I integrate the original project and develop a visual system interface. The example effect is as follows:
The official open source APK project available on the Android side, the address is here , as follows:
If you are interested, you can play with it.