CCTV 315 this year, face recognition vulnerability was ripped open.
Face recognition "Defense" was torn
The host provided a photo of his own on-site photo and was processed on site, quickly creating a 3D face model that was exactly the same as the viewer himself. Subsequently, the host will be the audience's 3D face mask directly on his face, "face" & rdquo ;, in accordance with the APP in vivo detection prompts to complete the blink, turned around, smiling and other activities, even deceived the APP, successfully completed In vivo testing and certification.
For a time uproar in public opinion.
In fact, such as the use of "video" crack face recognition equipment news is not uncommon. People use some 3D modeling technology can drive the user's single photo or video to make the system requirements of shaking his head, mouth, blink and other actions.
So, looks tall and convenient face recognition technology really safe? In the case of a person without knowledge, whether others can easily through the candid video or access pictures, using this technology to imitate the faces of other people, and then successfully achieve the face unlock, break the face recognition identity verification?
Lei Feng network learned that at present, face attack techniques are commonly attacked there are two main ways: First, the paper remake, that is, by printing the user's photos to verify; Second, all kinds of screen video remake and synthesis.
For the above two points, many technology vendors are based on the face recognition program did some in vivo detection protection (mainly shaking his head, blinking, etc.). However, it was later discovered that none of these technologies exceeded the ideal safety ceiling because technological advances were still more rapid than those of human factors.
This shows that, relying on the original pure 2D self-timer photos, has no small security limitations. Analysis of the human face 3D information, physical texture, and even tap more modal sensor information, is the only way to achieve safe brushing.
& ldquo; Light living & rdquo; on the line, to fill in face recognition security loopholes
At the Tencent Worldwide Partner Conference in 2017, Tencent Youtu introduced a ray of living technology called Aurora Guard for the "Brush Face" vulnerability. It emits light of different colors and intensities through the screen, projects onto the face and is received by the camera, and finally processes and convolves the neural network using a series of algorithms that analyze and derive the 3D face and texture information.
It is reported that the living body detection technology can effectively prevent malicious users to use photos, masks, dummies and even video forgery to deceive the system's security risks.
Leifeng Network learned that there are also technology vendors through the acquisition of infrared light and visible light as input data, combined with hardware-specific application-driven features, through infrared imaging, three-dimensional imaging detection, infrared and visible imaging matching identification, which is based on a special hardware 3D method ; The other is to rely on pure 2D method, through the analysis of the human skin texture and minor movements caused by changes in the law of optical flow in vivo testing.
On the surface, Tencent gifted graph "light living" compared with another major technological breakthrough. From the very beginning, the technology eliminates the need for sophisticated custom infrared spot emitters as the light source, opting for the most straightforward light source on mobile devices to emit variable light: your screen. The screen emits light of different colors and intensities, projects onto the face and is received by the camera. Finally, through a series of algorithms and rolls and neural networks, the person needed for living judgment is analyzed and derived from the video containing the changing light signal Face 3D and texture information.
Although the recent introduction of many manufacturers of infrared living technology can temporarily curb the face recognition device is "cracked" phenomenon, but due to larger device modules, most are still applied to TO B or TO G scene, not suitable for TO C scene Use, while the selection screen emits variable light as a light source & ldquo; Light living & rdquo; technology can be adapted to all mobile devices and brush face machine. If the technical stability is good, it is foreseeable that the equipment loaded with "Light Living" technology will soon be launched and popularized, making face recognition truly safe, efficient and rapid.
"Light living" technology really feasible?
General Assembly introduced, Tencent gifted newly opened living technology in the light of the security performance has exceeded the conventional technology, the calculation of 3D and texture information is only one of the purposes; the other is to use an instant generated random verification code encoded into the light Signal changes, in addition to analysis of the subsequent video analysis 3D information, but also the synchronization of the light changes in the screen signal decoding, test whether the code match, the formation of a verification code without user interaction to participate in the closed-loop verification process, Security level up by one level.
It is not difficult to see that Tencent gifted Figure is accelerating to create a face verification scenario "security barriers". Compared to the previous simple living and original lip language live body, loading the upgraded light living technology has four major advantages:
1, high security. Able to fully prevent electronic images, photos, videos (including video remake and video synthesis, etc.), masks and other attacks;
2, good user experience. Without any action with just a brief hold gesture, you can instantly complete brush face validation, truly unaware of recognition.
3, recognition speed. In practice, the fastest in vivo detection can be completed within 0.7 seconds to verify.
4, without custom hardware. Universal platform and model support for all mobile devices and brush face machine use.
From a theoretical point of view, perhaps this is a substantive breakthrough. In fact, there is a big difference between the light reflected by the screen and the light reflected by the face. The real face When collecting data in front of the device, the resulting reflection is totally different from the picture displayed on the screen. Neural network based on massive data in the device can capture small differences in the video sequence at random, to determine whether the measured object is a real or a malicious attack.
It is understood that the excellent map light living technology is currently in the "micro-public banks" application, so you can rest assured that "Brush Face" r.
Judging from the past cases, for face recognition, dynamic synthesis technology is indeed a potential means of attack, Tencent gifted Tu on-line "Light living" technology is indeed able to effectively prevent it.
Dialectically, absolute security equals unlimited costs. AES encryption can also be quickly cracked, we have to face positive 315 events this year and follow-up a variety of face recognition was compromised news. Like all security industries, face recognition has the spear and the shield. Attacks and attacks have always been a spear and shield. They have been playing games and escalating constantly to make the technology develop in a better direction.