You must remember that at the opening ceremony of the Consumer Electronics Expo in Hanover, Germany in 2015, Ma Yun, Chairman of Alibaba’s Board of Directors, showed the face recognition technology to the outside world for the first time. He placed his face in the recognition frame of the machine, and the system automatically recognized and completed the payment, and bought him a commemorative stamp of the Hanover Industrial Expo in 1948.
Nowadays, the scene of facial recognition technology has moved from Hanover, Germany to a broader field. At present, face-brushing payment, railway station entrance, candidates’ admission, etc. will be applied to face recognition technology, and the era of face-brushing has arrived.
Since Apple integrated fingerprint identification on iPhone5s, biometric identification has really become popular, and Apple’s latest iPhone has cut off fingerprint identification and used facial recognition as the only biometric identification method.
What followed was a series of follow-up by other manufacturers. After all, Apple’s facial recognition integrated on mobile phones is closer to the facial recognition recognized by the public. It is already a joke that the initial facial recognition can be unlocked with photos, and the recognition speed is even worse.
However, with the popularity of fingerprint recognition today, is it necessary for smart locks to follow up with mobile phone manufacturers and add facial recognition to smart locks to completely replace fingerprint recognition? The author thinks it is necessary to consider this problem from three angles: 1. Rejection rate; 2. False recognition rate; 3. Identify motives
It should be noted that up to now, Apple’s "face ID" recognition method is the most advanced, and it is also the most likely facial recognition scheme to be adopted on a large scale by future smart locks, so the following facial recognition is based on Apple’s "face ID" as an example.
First, the rejection rate
Rejection rate and false recognition rate are two important indexes in fingerprint identification. The rejection rate, that is, FRR(False Rejection Rate), is simply the probability of giving the correct information (such as fingerprints and faces) the wrong rejection.
Shortly after Apple’s iPhoneX went on the market, many people complained that the frequency of using passwords to unlock mobile phones was much higher than that of all previous mobile phones using fingerprint recognition. Many people also complained that they could not unlock them by facial recognition when lying on the bed, and they could only enter the password to unlock them. This is the result of the high rejection rate of facial recognition.
But in fingerprint identification, this problem rarely happens. This is the difference between facial information and fingerprint information.
Fingerprint identification is to press the finger on the identification module, but the fingerprint image pressed on the plane usually does not change greatly, and the image features are stable and the information is dense, which makes the accuracy of identification after encryption more stable.
However, the features of face are different. Although "face ID" has evolved from 2D image recognition to 3D depth of field recognition, a neural network chip has been added to the processor. However, the facial features of each person are mainly formed according to the bones, which will change in different situations.
For example, let a person run on a treadmill, and then record the facial information with a video recorder. Except for the shaky expression pack, you will find that the facial feature information of almost every screenshot will be different, and the two screenshots with big differences will even make you think it is two people. Humans sometimes can’t recognize a person’s facial information in different situations, let alone the neural network chip that has just started.
The above situation does not include makeup, glasses, masks, dark circles and bags under the eyes caused by poor sleep. As for getting fat and thin, Apple claims that it can gradually adapt to the change process, while other 3D face recognition technologies do not seem to have achieved this effect, but this technology can follow up soon.
In addition to the rejection rate, the false recognition rate is also a controversial place in facial recognition.
Second, the rate of false recognition
False recognition rate, that is, FAR(False Rejection Rate), is also called "false recognition rate". It is to judge the wrong information as correct and finally pass the authentication. It is difficult for this indicator to have any problems in fingerprint identification, because it is difficult for you to find two people with similar fingerprints and the characteristics of fingerprints are irregular. Therefore, in fingerprint identification, the high false recognition rate is most likely related to encryption algorithms.
In facial recognition, this false recognition rate has a big problem. The most prominent problem is the similarity of faces caused by twins. Not long ago, someone did this experiment. One of several pairs of twins entered facial information through iPhoneX, and then the other twin tried to unlock it.
Judging from the actual test results, the false recognition rate in this case is as high as nearly 50%. It is even more news that two people who are completely unrelated in China can unlock the same iPhoneX just because they look similar. Although the feature points detected by the "face ID" on the iPhoneX are far more than the features collected when human beings observe the face, they will still be defeated in front of twins, especially identical twins who look very close.
Many people (including the author’s face blindness) often can’t distinguish between two twins, and it takes a long time to evolve the algorithm to make the neural network chip that is not mature enough to distinguish twins or two people who look similar.
Fortunately, however, with the evolution of facial recognition from 2D image recognition to 3D depth-of-field recognition, the situation of unlocking the mobile phone with only photos or videos will no longer exist, which is also the advantage brought by the progress of recognition technology, and also directly improves the false recognition rate of facial recognition by several levels.
However, the mistake of facial recognition is put on the smart lock to eliminate the problems caused by the bad relationship between very few twins. Most twins will welcome their brothers into their homes. Therefore, in the face of mature 3D depth-of-field facial recognition, the recognition rate of smart locks should not be a big problem. After all, the difficulty of recognition rate is not to be cracked, but to find the person who looks like it.
The real big problem is what to say next: identifying motives.