At the opening ceremony of the 2015 consumer electronics Expo in Hannover, Germany, at the opening ceremony of the Hannover Consumer Electronics Fair in 2015, Ma Yun, the chairman of the board of directors, presented the face recognition technology to the outside world for the first time. He placed his face in the machine's identification frame, automatically recognized and completed the payment, and bought him a 1948 Hannover industry. Commemorative stamps of the Expo.
Today, face recognition technology has gone from Germany to a broader field in Hannover. At present, brush face payment, railway station entry, and admission of candidates will be applied to face recognition technology. The era of brush face is coming.
From the beginning of Apple's integrated fingerprint recognition on iPhone5s, biological detection and recognition began to become really popular, and Apple's latest iPhoneX cut the fingerprint out of the fingerprint, using face recognition as the only way of biometric recognition.
What followed is a series of follow up from other vendors, after all, Apple's face recognition on the mobile phone is closer to the recognition of the face. The initial facial recognition can be unlocked with photos, which is already a joke, and the recognition speed is even worse.
But today, with fingerprint recognition, is there a need for smart locks to follow up handset vendors and add facial recognition to intelligent locks instead of fingerprint identification? The author thinks that it is necessary to consider this problem from three angles: 1, rejection rate; 2, false recognition rate; 3, recognition motivation.
It needs to be explained that, up to now, Apple's "face ID" is the most advanced and most likely to be widely used by future smart locks, so the face recognition below is exemplified by Apple's "face ID".
One, refusing to be true
Rejection rate and false recognition rate are two important indexes in fingerprint recognition. False Rejection Rate (FRR) is simply the probability that the correct information (such as fingerprints, faces) will be rejected incorrectly.
Soon after Apple's iPhoneX was on the market, many people complained about the frequency of unlocking passwords on their phones, much more than all the previous use of fingerprint recognition phones, and many people complained that when they were lying on the top of the bed, they could not be unlocked by the face recognition, only to be lost in the password to unlock. This is the result of the fact that facial recognition is too high.
This problem rarely occurs in fingerprint recognition. This is the difference between facial information and fingerprint information.
Fingerprint identification is to press the finger on the recognition module, and the fingerprint image on the plane is usually not changed greatly. The image features are stable and the information is dense, which makes the accuracy of the recognition more stable.
But the features of the face are different. Although "face ID" has evolved from 2D image recognition to 3D depth recognition in obtaining information, a neural network chip is added to the processor. But the facial features of each person are mainly formed according to bones, and they will also change in different situations.
For example, let a person run on a treadmill, record the following information with a VCR, and you will find that the facial features of almost every screenshot are different except that the facial features of almost every one of the screenshots are different. The two screenshots of a big difference may even make you think two people. Humans sometimes can't recognize a person's facial information in different situations, let alone the nascent neural network chip.
And the above mentioned situation, has not included makeup, wearing glasses, wearing a mask, poor sleep caused by dark circles and bags under the eyes and so on. As for fat and thinner, Apple claims to be able to adapt gradually in the process of change, and other 3D - face recognition technologies seem to have not reached this effect, but the technology can soon follow.
In addition to repellent rate, recognition rate is also a controversial area of facial recognition.
Two, the rate of false recognition
The false recognition rate is FAR (False Rejection Rate), also known as "false recognition rate". It is to judge the wrong information correctly, and finally pass the authentication. It is difficult to find any problem on fingerprint recognition, because it is difficult to find two people with similar fingerprints, and the characteristics of fingerprints are not regular, so on fingerprint recognition, the high false recognition rate is most likely to be related to the encryption algorithm.
In face recognition, this recognition rate is very big. The most prominent problem is the similar appearance of twins. Not long ago, someone did the experiment, recording one of the twins' iPhoneX into the face information, and then trying to unlock another of the twins.
Judging from the actual test results, the false recognition rate of this situation is as high as nearly 50%. There's news that two people in China who are completely unrelated can unlock the same iPhone X simply because they look alike. Although the features of the "face ID" test on iPhoneX are much more characteristic than the human observation of the face, the twins, especially the identical twins, which are very close to the face, still lose the array.
Many people, including the writer's face blindness, often don't distinguish between two twins, and a neural network chip that makes the algorithm less mature enough to separate twins or two people with similar looks will take a very long time to evolve the algorithm.
However, it is gratifying to see that face recognition has evolved from 2D image recognition to 3D scene.