Original Source Here
Why does 3D face recognition technology make “face swiping” safer?
After visiting the supermarket, holding shopping bags in hand, and looking at the terminal device at the cashier, you can pay by swiping your face, saving everyone’s waiting time in line. Such a convenient experience is due to the popularity of face recognition technology.
However, while “face-swiping” brings convenience to our daily life, its safety has attracted more attention. Not long ago, the researchers from Real AI cracked the facial recognition unlocking system of 19 Android phones by adversarial sample attacks. Only one iPhone 11 hasn’t been cracked.
Why is face recognition system so easy to be cracked?
Briefly, current face recognition system can be divided into 2D plane and the 3D stereoscopic image recognition. According to the security level, 3D face recognition has a higher security level than 2D face recognition.
2D face recognition uses a camera to obtain an RGB color image of face, and then monitors the face on the image. The system will identify the eyebrows, eyes, nose, etc. on the face and extract features. Finally, the system will output a numerical string to compare the information in database, thus face identification process is finished.
With the emergence of more and more “bugs” in 2D face recognition, many companies turn to 3D face recognition.
Apple iPhone X is a familiar application of 3D face recognition technology. More and more Android phones, such as Huawei Mate 30 Pro, OPPO Find X, Xiaomi 8 transparent exploration version, etc., also use similar 3D face recognition technology.
3D face recognition is based on RGBD for multi-modal (RGB+depth map) image recognition or directly on 3D point clouds face recognition. After recognizing the three-dimensional coordinate information of each point in the space of face, the complete three-dimensional image is calculated and restored. So even if your head is not facing the camera, devices using 3D face recognition can also recognize you.
Take 3D structured light by iPhone X’s Face ID as an example. Its deep-sensing camera module includes an infrared lens, flood light sensing elements, distance sensors, and dot-matrix projectors. When working, the dot matrix projector projects 30,000 infrared point light sources that are invisible to the naked eye. An infrared photo is taken by infrared lens, and the depth of field information of the face is analyzed based on the dot matrix displacement on the photo. Thus, a 3D model will be synthesized.
The information obtained during the entire processing of 3D face recognition is greater, and the upper limit of recognition accuracy is much higher than that of 2D face recognition. At present, it can even be commercially used in scenarios with higher requirements such as face swiping payment.
As AI data service provider, Datatang has developed “ 1,078 People 3D Faces Collection Data”, “10,000 People 3D Face Recognition Data” and “ 1,000 People–3D Living Face & Anti Spoofing Data”. Datatang strictly abides by relevant regulations, and the data has been authorized by the person to be collected.
1,078 People 3D Faces Collection Data
The staff used Realsense SR300 to collect 16 segments of multi-illumination and multi-posture video for each person. The video resolution is 1,920*1,080. The gender, age, scene, action, glasses, distance, etc.is annotated. The data has been desensitized and doesn’t contain sensitive information. The accuracy of facial motion data exceeds 97%, and the accuracy of labeling is higher than 97%.
10,000 People 3D Face Recognition Data
The data was collected by iPhone X and iPhone XR, and 24 photos were collected from each person. The data covers a variety of actions, such as silence, head up, head down, head left, head right and with masks. The race, gender, age, scene, actions and lighting conditions is annotated. The accuracy of the action is more than 97%, and the accuracy of labeling is higher than 97%.
1,000 People–3D Living Face & Anti Spoofing Data
The data is collected by iPhone X and iPhone XR. The data covers multiple expressions, multiple face poses, adversarial samples, multiple lighting conditions, and multiple scenarios. The ID, race, gender, age, facial action, collection scene, and lighting conditions is annotated. The action accuracy exceeds 97%.
The original intention of technology empowering “face recognition” is to bring convenience to people’s lives, not to “run naked” without personal privacy. While the law continues to delineate the “red line” for face recognition, the industry also needs to establish technical standards for face recognition, design and develop mature face recognition solutions.
If you need data services, please feel free to contact us email@example.com.
Trending AI/ML Article Identified & Digested via Granola by Ramsey Elbasheer; a Machine-Driven RSS Bot