Before entering self-service banking, the camera will automatically recognize you and open the door with “brush face”; you can enter the password before standing at the ATM and then “brush face” in front of the camera to successfully withdraw money. This bridge that looks like a science fiction movie is increasingly being used in reality.
In April 2015, Ma Yun showed his face to the public at the IT Expo in Hannover, Germany. Last month, he claimed that the world's first ATM with face recognition was born in Hangzhou. After the adoption of face recognition technology in Guangzhou's pension certification, the subway will also try the technology to identify suspects. Even the WeChat circle of friends has been screened by an age-tested app.
Is the era of "brushing face" really coming? As a biometric technology, why is the “brushing face” so hot? Some Chinese Academy of Sciences experts believe that compared with other biometric technologies, face recognition technology has lower acquisition cost and higher convenience, but the disadvantage is that the reliability and stability of information is weak.
Brush your face to withdraw money: from April Fool's Day to joke
"The era of empty wages is coming." In the video of 1 minute 54 seconds, the young man with glasses said confidently at the camera. On April 1, 2013, Alipay official Weibo released the above video. The video called "empty payment" looks like a plot in a science fiction movie: users don't have a wallet, a cell phone, brush their face directly at the store, tattoos or even pets can pay. Subsequently, Alipay official confirmed that this is just a joke on April Fool's Day.
But only two years later, this joke was staged in reality. In May, the world's first A T M machine with face recognition was born in Hangzhou. The aircraft was jointly developed by Tsinghua University and a company. It is the first financial security device in China with independent intellectual property rights. The so-called A T M face recognition is performed by reading the photo information on the user ID card and then comparing the photo with the face of the payee recorded by the top camera of the ATM machine. It can be networked with banks, public safety and other systems to curb crime.
Not only can you identify the suspect, but you can also prove that "you are you"
Not only can the funds be obtained through the face A T M, but the Guangzhou Metro will use face recognition technology in the future. Recently, a face recognition system called “Future-oriented” developed by the First Research Institute of the Ministry of Public Security has been officially released to the outside world. After collecting facial information in the terminal, the system compares with the fugitive in the database to identify the suspect. Tian Qing, a researcher at the institute, said in an interview with the media that in mid-June 2015, the product will be commissioned at the Guangzhou subway station.
On April 27, the Guangzhou Municipal Bureau of Human Resources and Social Security released a message saying that Guangzhou developed a contactless face recognition technology. After the retirees completed their first facial data collection modeling, they only swept their faces in places where there was an Internet and a camera. You can complete the annual certification. In order to prevent pensions from being taken over, retirees must hold a certificate of survival each year to confirm eligibility. According to the new method, retirees must first carry out data modeling and collection, carry the second-generation ID card and other documents, go to the town and town retreat department to carry out facial data collection modeling, and through online qualification certification, you can directly operate online, greatly saving Time.
Financial payment and security industry has the most promising applications
"In the future, this market will be very large, and the future will enter a period of barbaric expansion." Zhang Hao, head of Shanghai Comic Technology Co., Ltd. told Nandu reporter. He believes that facial recognition technology will gain more imagination in the future of the security industry. The company developed the "Reconstruct and Identify Low Resolution Small Images" technology to help the public security department resolve the case. “In some cases, the suspect wears sunglasses to cover half of the face, and our technique can narrow the range. In addition, the pixels in the video are very low. From this software recovery, it can range from 500,000 to 600,000. Suspects In the range of a few hundred people." Zhang Wei said.
In addition, in major conferences, participants can upload identity information and enter it into the access control system to identify whether they are themselves. Shanghai's singular technology was used in the 2008 Beijing Olympic Games and the Shanghai AsiaInfo Summit in 2014. During the Beijing Olympics, the audience entered the Bird's Nest, in addition to the tickets, they also had to take photos one by one in front of the entrance channel. The camera captures the face in two seconds, locates the key points of the face, and extracts the features, then uploads the authentication results to the computer at the same time, and then compares the identity information of the viewer with the computer.
In addition to the security industry, another major application scenario for face recognition technology is in the field of financial payments. The fast-growing Internet finance industry needs the support of the personal credit information system, and the industrial demand is growing. User identification becomes the first step in the login credit system. This is the potential for large-scale commercial use of face recognition technology.
In April 2015, Ma Yun at the IT Expo in Hannover, Germany, demonstrated the face payment in public. This technology, developed by Ant Financial and Beijing Defiance Technology Co., Ltd., plans to replace the traditional password by “sweeping face” in the payment authentication stage after shopping. An ant Jinfu, a person in charge, said in an interview with the media that the face-sweeping payment will be used in Alibaba's global payment and account opening certification business.
Stronger than human recognition, but temporary aid
But brush your face, can you really make a safe payment? Sun Zhenan, a researcher at the Institute of Automation of the Chinese Academy of Sciences, believes that the current face recognition technology is not perfect. "If you leave the password input, it is impossible to use face recognition technology for deposit and withdrawal operations." Zhang Xi also said that he believes that in the banking field, face recognition technology can only be used as an auxiliary means. "The bank's requirements are zero tolerance and no problems can arise."
Even so, "Face recognition technology is stronger than human recognition, and it is much stronger," said Xie Yinan, general manager of Beijing Defiance Technology Market and Operations Department, to Southern Reporter. "For example, the accuracy of face recognition by a bank teller may reach One millionth of the misunderstanding rate, the pass rate can exceed 90%; and our best score is one hundred thousandth of the misunderstanding rate, the pass rate can exceed 97%-98%." Biometric technology has been obtained in the world The importance of big technology companies. Apple launched fingerprint recognition T ouchID, Samsung, Xiaomi has begun to test the water face unlock screen
principle
Divide the face into more than 100 points to simulate the neural network operation process
Face recognition research began in the late 1960s and early 1970s, and it is a kind of biometric technology with fingerprint recognition, palmprint recognition, and iris recognition. The early face recognition technology is to measure the geometric relationship between the face of the person's face according to the corners of the eyes, the nostrils, the mouth and the chin. Through the face template in the image library, the degree of similarity with the face to be recognized is grayed out. Face recognition. The drawback is that it is easy to lose useful information, and the recognition ability is poor in the case of changes in perspective, expression, and the like.
In the 1990s, with the rapid development of computer technology, after several iterations of the technology, face recognition technology has evolved from the initial identification of a single grayscale image of the background to the identification of different sides. Static face, currently able to perform face recognition dynamically.
Internationally, the leading research institutions are the Massachusetts Institute of Technology in the United States and the University of Cambridge in the United Kingdom. Domestic research on automatic face recognition began in the 1980s. The main research units are Tsinghua University, Harbin Institute of Technology, and Chinese Academy of Sciences. In recent years, under the wave of Internet entrepreneurship in China, a number of new biometrics entrepreneurial teams with academic backgrounds have begun to emerge, such as the face++ team under the contempt of technology and the excellent map team of Tencent, which has adopted more advanced “deep learning”. The algorithm uses a neural network to simulate a human neural network operation process, which can achieve higher precision.
Founded in 2012, the face+ + team has partnered with Ant Financial and a number of banks. Xie Yinan, general manager of the marketing and operation department of the company, introduced the reporter to Nandu. "For example, someone in front of the camera grabbed half of his face with his hand. According to the traditional regression algorithm, the system thinks that he is not a human face. Because it does not It conforms to the standard structure of the face obtained by the regression operation; and the deep learning algorithm, on the basis of judging the face, will enter the second level of judgment and multiple verification."

As a new stage in the research of artificial intelligence technology, the deep learning algorithm is equivalent to "teaching the machine to imitate the learning mechanism of human beings". With this algorithm, the operator can provide the system with a large amount of data to enable it to learn. For example, the deep learning algorithm robot, by scanning the picture of the cat on the Internet, the operator enters the "cat", and after a period of time, the machine associates the furry little animal with the "cat" and can identify itself as a cat. Xie Yinan introduced that the company's face verification technology, "the first step is to prove that you are human, the second step is to prove that you are you." Under the deep learning algorithm, the system can identify faces more intelligently.
Because of time, there are often some differences between the ID card photo and the user's current photo, such as the width of the face, but the distance between the pupils is relatively constant. Traditional regression operations may have errors, and deep learning algorithms divide the face into more than 100 key points to avoid errors.
Compared
What are the advantages and disadvantages of “brushing face” compared to other biometric technologies?
Sun Zhenan, researcher at the Institute of Automation, Chinese Academy of Sciences: The advantage of face recognition technology is that the collection is relatively intuitive, the cost is low, the convenience is high and the user is more acceptable. But its shortcoming is that the reliability and stability of the information is weak.
Specifically, the amount of information contained in a human face is relatively small compared to biological features such as fingerprints and irises, and the complexity of the changes is insufficient. For example, if the fingerprints or irises of two people are basically the same, it may take several tens or even hundreds of bits (the unit of measurement of the amount of information) to be completely coincident. But if it is a human face, a dozen or so bits can be coincident. In the world, many faces with similarities can be found. Therefore, the recognition of the face is not very high, it is not so unique.

In addition, the internal changes of the person and the changes in the external environment will affect the information stability of the face at the time of collection. Compared with the previous face recognition technology, the current face recognition technology has improved, but the specific application still can not achieve perfect state. For example, the face recognition technology used on the A T M machine is an auxiliary authentication function based on the use of the password information. If you leave the password input, it is not possible to use face recognition technology for deposit and withdrawal operations.
The president of a technology company: According to the precision of identification, it is true that the recognition accuracy of iris, fingerprint and face is reduced in turn. However, the face can be raised according to the improvement of the camera, and this rising space is large. From the point of view of applicability, you now let all Chinese people extract iris information and fingerprint information. This is difficult and unrealistic. And each of us has a photo of the ID card. From the perspective of the comparison library, face recognition has certain advantages. Another problem is the device. Iris recognition, fingerprint recognition requires additional equipment, and face recognition, as long as there is a mobile phone. From the perspective of large-scale promotion, the bottleneck of face recognition is relatively small.
Dispel doubt
A. If people get older, can they still be identified?
Sun Zhenan: The current face recognition technology is acceptable for facial features changes within 2 to 3 years. But after a longer time span, the accuracy of face recognition will be less reliable.
The head of a technology company in Shanghai: young people under the age of 15, the face is developing. And people over the age of fifteen can be identified within ten years.
B. How does the current technology distinguish between twins?
Xie Zong: Twins, in fact, the machine is better than people. We look at the twins, the face is quite like, the nose is quite like, this is a perceptual knowledge, and the machine uses a logical algorithm to give a similarity. Of course, you may say that some twins are very similar. It requires a more detailed camera. We can divide the face into more and more key points. If the texture of the skin can be seen clearly, there is no problem.
But at present it is too close, we can't do it. For example, the face is exactly the same, the distance from the eyes to the squat is exactly the same, even the aunt is difficult to distinguish, and now we can't do it.
C. Can you recognize it after plastic surgery? What if someone has a heavy makeup?
Xie Zong: After the plastic surgery, you should change your ID card first. Because what we have to do is that you and the photo on the ID card are the same. If technology is used in the financial sector and it is necessary to control risks, this situation will certainly not pass. But if it is micro-shaping, such as face-lifting, we can identify this.
Make up the makeup... It depends on what he does. If you go to a financial institution to open an account, the auditor of the other party will also ask you for things, such as not wearing glasses or wearing a beautiful woman. As long as it is a simple light makeup, it is not super heavy makeup, smoky makeup, and does not damage the characteristics of the face itself, we are all recognizable. In addition, we can identify the glasses. For example, someone wearing a pair of black-rimmed glasses can identify this.
D. Will the printed photos be confused?
Xie Zong: Our verification technology is now in two steps. The first step is to prove that you are human, and the second step is to prove that you are you. Be sure to stand in front of the camera is a person, not a photo, this is the first step to prove that you are human. We ask each other to make three interactive videos in a row. The algorithm monitors in real time, can observe the subtle movements of each observation point, and then judges this. For example, for a normal person, the degree of his eyelid fit should be, we have a set of rules. If he does not meet the rules, he will not go.
In the world, many faces with similarities can be found. Therefore, the recognition of the face is not very high, it is not so unique.
— Sun Zhenan, Research Fellow, Institute of Automation, Chinese Academy of Sciences
You now let all Chinese people extract iris information and fingerprint information. This is difficult and unrealistic. And each of us has a photo of the ID card. From the perspective of the comparison library, face recognition has certain advantages.
— Wang Gaoxin, President of Chongqing Huifan Technology Co., Ltd.
Facial Time Attendance
|
HF-RA05 |
HF-RA07 |
Dimenson |
195*85*32mm |
218*120*25mm |
OS |
Android 8.1 with Octa-core 1.5GHz CPU, GPU support |
Memory |
2GB RAM and 8GB ROM |
Screen |
5" IPS Screen; 800*480 resolution |
7" IPS Screen; 800*480 resolution |
Face Capacity |
1000 standard; optional 2000/5000 |
10,000 |
Log Capacity |
10,000(auto override) |
Face Verification |
Single, and multi face |
Detecting Distance |
0-4 meters, adjustable |
Cameras |
RGB Camera & Infrared camera |
Communication |
RS-232 & RS-485; TCP/IP; Wifi |
Verification Time |
<0.7 s |
Muliti-face verification |
5 faces |
You can click here for more information about Face Recognition Detecting Device
For the parts whose faces are not easy to change, calculate their distance, angle and size in the facial features, and form numerical data for comparison; the most inaccurate and most vulnerable to all biometric techniques, the location of the user's face and The surrounding light environment can affect accuracy.
odor
The special odor determined by the human body is extremely stable; it is used more in criminal investigation.
voice
It is not a simple sound comparison, but the collected sound is converted into a numerical value, which is a comparison between the values; when the cold, pharyngitis and vocal cords are injured, the sound is very different from usual, and the error is also large; The possibility of camouflage and control your own voice.
Fingerprint Module |
HF7000 FBI Cerificated Bluetooth Fingerprint Scanner |
Fingerprint Sensor |
Live Finger Detection |
Image DPI |
508 DPI |
Imge Resolution |
256*360 |
Capture Size |
18*12.8mm |
RFID Card |
NFC Card |
Support OS |
Windows, Linux, Android and OS |
Fingerprint Capacity |
1000 templates |
Battery |
Built-in Li battery, 3.7V 1000 mAh, working 8 hours |
Connection |
Bluetooth/usb |
Fingerprint Template Standard |
ISO, ANSI, WSQ |
The fingerprint of each person is different, that is, the fingerprint fingerprint of the same person is also obviously different. Therefore, by comparing the detailed feature points of different fingerprints, it can be used for identification and application. However, some people have fewer natural fingerprints and it is difficult to collect images. The environment is very demanding when collecting, and it is sensitive to the condition of the fingers. Scars, peeling and even old cockroaches will affect the effect. It has been reported that many office white-collar workers use the fingerprint film to pass the fingerprint attendance machine.
DNA fingerprinting
It refers to the DNA polymorphism with complete individual specificity. It is the most unique feature in DNA. The accuracy of identification using DNA fingerprinting is as high as 100%. It is mainly used for genetic analysis of diseases, paternity testing, criminal investigation, etc., and sampling is convenient, but the fee is charged. Higher, not yet widely promoted.
Vein Scanner
The image of the finger and the palm vein is collected by the camera, and the feature is extracted from the digital image and stored in the database. Because the veins are hidden inside the body, it is difficult for outsiders to contact, and it is difficult to forge. At the same time, the veins are basically unaffected by physical conditions and external factors, and the anti-interference is very good, and the recognition rate is very high. Foreign countries have precedents for access control and withdrawal.
Iris Reader with Smart Card
Verifiacation |
IRIS, face, fingerprint and RFID |
OS |
Android |
Function |
Attendance and Access Control |
Access Control |
Time zone, grouping, Anti-passback |
Battery |
Built-in |
POE |
Yes |
Screen |
Touch Screen |
Face |
Moving face identification |
Operation Temp |
-10 ℃ - 50℃ |
Verification |
within 1 second |
Power supply |
12V 3A |
SDK |
Yes |
The human iris matures around the age of two and remains stable afterwards, unless the eye surgery and the brain are severely damaged. In most cases, the iris remains unchanged for life and the recognition accuracy is very high. However, this technology is powerless for blind and eye patients; the collection equipment is expensive and difficult to promote; and it is currently difficult to identify black eyes.
Palm print & Face Fingerprint Time Attendance
Biometric Measurement
|
Palm Vein & Fingerprint
|
User capacity
|
5,000
|
Palm capacity
|
2,000
|
Face capacity
|
5,000
|
Fingerprint capacity
|
10,000 (upgrade to 30,000)
|
Log capacity
|
200,000
|
CPU
|
1GHz
|
LCD display
|
4.3" Touch color screen 800x480 pixel or 4.3'' color screen
|
Interface
|
GPRS (4G, 3G) ,TCP/IP, USB (host and slave), 1 × Relay, wiegand in & out
|
Language
|
English, Arabic, French, Korean, Spanish, Portuguese, etc
|
Attendance mode
|
Palm/Face/Fingerprint/Card/Password; Fingerprint+Face/Palm; |
Card+Face/Fingerprint/Palm; Password+Face/Fingerprint/Palm
|
Voice instruction |
16-bit Hi-Fi voice & sound indication |
Verification time |
< 0.5 second [1:N] |
FRR (False Rejection Rate) |
< 1% |
FAR (False Acceptance Rate)
|
< 0.0001%
|
Power source
|
12VDC/2A
|
Operating temperature
|
0℃-45℃
|
The image of the palm from the end of the finger to the wrist is collected and controlled by genetics. Even if the palm is healed, the palm line grows as it is and has stability.