Recently, "brushing the face" has become a hot word, and face recognition technology has continuously entered the public view. Apple's new iPhone X has a "face brushing" unlocking function that can be applied to Apple Pay and various apps that require authentication; the commercial pilot for the first "face brushing" payment is also open at a KFC restaurant in Hangzhou; some banks are trying to make it work. Automatic cash machine "brush face" withdrawal function; high-speed rail ticket check, hotel occupancy is also using "brush face" technology ... ...
Face recognition has been applied in all fields of people's clothing, food, housing and transportation, ushering in the "blowout period" of use, among which the financial and security industries have become "pioneer" applications. With the continuous expansion of commercial scenes of face recognition technology, the market has great potential, and capital is smelling business opportunities. According to the data from the Prospective Industry Research Institute, the market size of the face recognition industry in China has exceeded 1 billion yuan in 2016, and is expected to reach about 5.1 billion yuan by 2021.
The "Brush Face" era brings a huge market
Face brushing, face brushing, face brushing, face brushing... With the increasing maturity of face recognition technology, the era of "brushing the face" is coming. In the eyes of people in the industry, face recognition technology is constantly breaking through the "threshold" of application in various industries, bringing about increasingly rich application scenarios.
“With the deep learning algorithm debuting, face recognition accuracy has significantly increased compared to five years ago.†Yan Shuicheng, vice president of 360 company and director of the Institute of Artificial Intelligence, said that the information extracted from various devices to capture faces will be formed. With the data, the large amount of data that has been accumulated has become a "sufficient nourishment material" with perfect feeding technology.
Chen Jidong, head of biological recognition technology for ant gold suits, said that thanks to the rapid development of deep learning in recent years, we can make machines simulate the learning process of the human brain based on neural networks and train through convolutional neural network models and massive image data. . Biometrics has improved from the previous 70%, 80% accuracy rate to 99.6% or even 99.7% in the past two years, with commercial conditions. At the same time, the misrecognition rate of face recognition technology in payment scenarios has reached one in 100,000.
Xie Yinan, Vice President of Technology, told reporters that there are three major application directions for face recognition technology, one for 1:N authentication, to determine whether an entity is a member of a specific group, for personnel access management and urban security, including public security Captured fugitives, community access control brushing system, and VIP management of some businesses.
The other is 1:1 authentication, which proves that the information of the person and the certificate is unified, and is mainly applied to scenes requiring real-name authentication. China Southern Airlines launched this year in June in Nanyang Airport in Henan, “Brushing the plane boarding planeâ€, Wuhan Railway Station and Guangzhou South Railway Station launched the “brush face into the stationâ€, that is, this category.
The third type is a living body test, which proves that the real person is operating the business and then doing account authorization. CITIC Bank's ATMs and mobile clients can carry out remote authentication, Haitong Securities can open accounts remotely, and Didi platforms can check whether drivers are registered drivers.
Face recognition is slowly moving from the line to the line, unveiled in scenes such as unmanned retailing, quick payment, and hotel occupancy. Capital finds business opportunities among them. In July of this year, Shangtang Technology announced the completion of a $410 million Series B round of financing. Shanghai Yitu Science and Technology and Beijing Cishi Technology completed the Series C round of financing, amounting to 380 million yuan and 100 million US dollars, respectively. According to the data from the Research Institute of Prospective Industries, the market size of the face recognition industry in China has exceeded 1 billion yuan in 2016 and is expected to reach about 5.1 billion yuan by 2021.
Business applications continue to enrich
"Economic Information Daily" reporter learned that the application of face recognition technology in the financial sector has been explosive growth. The acceptance of the world’s first ATM with face recognition from China’s self-developed technology has led to the launch of “brush face payment†applications by internet companies Ant Financial, Jingdong, Suning, etc., and pilot face recognition applications by traditional banks such as China Merchants Bank. ... Applications in the financial sector such as payment, withdrawal, and loans have come before other fields of application.
In September of this year, the application of “brushing the face†in the financial industry made eyeballs. In Apple's new machine conference, iPhoneX has a “brush face†unlocking function as the focus of attention, Apple said that this feature can be applied to ApplePay; financial technology company Ant Financial and KFC jointly announced “brush face payment†into the commercial pilot At the stage, this is the time when the face-lift payment went from the line to the line, and it was the first time that it really fell into the consumption of the commercial scene.
In the KPRO restaurant of KFC in Hangzhou, Vientiane City, the "Economic Information Daily" reporter saw many consumers try to "brush face pay": choose a good meal in the buffet ordering machine, enter the payment page, you can choose to follow Alipay, WeChat mobile After the payment option, the new option “face payment†is performed, and then face recognition takes about 1-2 seconds, and then the mobile phone number bound to the Alipay account is input, and the payment can be made after confirmation, and the process takes less than 10 seconds.
In the area of ​​urban security where the degree of difficulty is extremely high, face recognition is also making a difference. In the past, face recognition technology could only handle hundreds of people's data comparisons, but now it has developed to more than 10,000 people and even higher-level data comparisons, and it is unfavorable for breakthrough shooting angles, complex light changes, and low resolution. Conditions to help the public security agencies quickly arrest fugitives.
The reporter was informed that face recognition company Contempt Technology has provided real-time alert data services for multiple public security systems, which directly assisted the police in cracking 1032 cases, and captured and controlled more than 2,000 fugitives. A Public Security Sub-bureau in Chongqing used the Shang Tang Technology's portrait matching system to identify 69 suspects within 40 working days, which is 200 times more efficient than manual labor.
Before the arrival of face recognition technology, biometric recognition methods such as fingerprint recognition and iris recognition have been widely used in life. However, respondents said that compared to the biggest advantage of face recognition is "non-contact", which can greatly increase the speed of the system response, improve ease of use, while avoiding fingerprints and other contact-type recognition of disease transmission and other health Hidden danger.
In addition, the "non-involved, non-intrusive" feature means that data can be collected without the user's cooperation, which is conducive to the application of public security in security and other fields.
Face recognition technology is also increasingly used for entertainment. The facial recognition unlocking function becomes a tablet “selling pointâ€, smart albums can classify photos by identifying faces, and “make-up†apps automatically recognize faces and “make-up†them... For example, a popular "FACEU" software , You can turn the user's picture into a picture such as Dasheng, Rabbit, etc., and interact with friends on social platforms such as friend circles and Weibo.
Insiders believe that smart homes will be one of the application scenarios for face recognition in the future. Intelligent security doors will only open when the owner is standing in the doorway. Smart TVs can identify who you are and push it to your frequently watched programs. Service robots can also provide corresponding services based on the identity of the object. In the future, face recognition technology will make it possible to dig deeper into user information. Businesses can analyze the purchase behavior of members and then arrange business layouts or promotional activities in a targeted manner.
Technical accuracy breakthrough can be expected
Experts believe that in the future, face recognition technology will continue to break through. On the one hand, accuracy and safety will continue to increase, and processing capabilities for special circumstances such as face lifts and twins are also improving. On the other hand, the magnitude of face recognition can continue to increase. When the technology has progressed to extract and match a face in a database of hundreds of millions of photos, the application scene will gradually expand.
According to Yan Shuicheng, usually face recognition involves the following steps: the camera or professional device first collects pictures, face detection technology locates faces in pictures, and then repositions features such as the corners of the eyes, nose, mouth, and contours of the face. Perform corrections including light compensation or culvert removal. Then use the deep learning algorithm to extract the identity features and compare it with the face features in the database to identify the face identity.
The industry believes that the key to the technology lies in finding out the relationship between the key points and the facial expression network on different facial images and finally determining whether these images are the same person. However, face changes, different angles, different makeup can affect the capture of key features.
In addition, the “brush face payment†is performed under the public equipment and under the open environment. The real scene is complex and changeable, and the security requirements are even higher. Does biometric technology bring more convenience or challenge to people's lives?
Doubt one: how to ensure the accuracy of "brush face"?
When measuring face recognition capabilities, many companies claim that their accuracy exceeds "99%." In this regard, Gong Yihong, an adjunct professor of the Xi'an Jiaotong University's School of Telecommunications and a national "Thousand Talents Program" professor who has long studied machine learning, said that the accuracy here refers to the results achieved in some world-famous face database comparisons, but in reality In use, this accuracy is greatly reduced.
Yang Fan, co-founder of Shangtang Technology, also believes that these accuracy levels are achieved under certain preconditions, but the real-life application scenarios are complex and changeable. The crowd samples are larger, and conditions such as different light, attitude, and resolution may give machine recognition. Bring difficulties.
However, this does not mean that the technology must be 100% accurate before it can be used. “There is no perfect technology in the world. Any technology is error-prone. However, if the accuracy of the technology can meet the requirements in a specific scenario, and the risks brought by mistakes can be sustained, then it is valuable. "Yan Shuicheng said.
Apple introduced the face of the new machine iPhoneX ID function using a dot matrix projector, infrared lens and floodlight sensor components composed of advanced original deep camera system, in the A11 bionic powerful dynamic support can be mapped and face recognition face. This function projects more than 30,000 infrared light spots that are invisible to the naked eye, and then transmits the resulting infrared image and dot pattern to the neural network to create a mathematical model of the user's face. The data is then sent to a safety compartment to confirm Whether the data matches. Moreover, the appearance of the user changes over time, and the technology can be adjusted accordingly.
Ant gold clothing introduction, Alipay in the KFC's ordering machine is equipped with a 3D infrared depth camera, in front of face recognition, through a combination of hardware and software for live detection, to determine whether the collected face is a photo, video Or the software simulation is generated to avoid identity fraud caused by all kinds of face forgeries.
Doubt two: Can twins, over-makeup, and cosmetic surgery be resolved?
“The angle, light, expression, age, make-up, occlusion, photo quality, etc. of people’s faces will affect our judgments, and as the size of the database increases, the probability that two different people look like it will rise rapidly.†Chen Jidong proposed Biometrics is facing difficulties, but he believes that deep learning will make computers smarter and overcome these difficulties.
Yan Shuicheng said that in the face of special circumstances such as twins or before and after the plastic surgery, whether or not the machine can be identified depends on the specific circumstances. For example, if the size of the cosmetic surgery is too large, it is possible that the machine cannot be identified. In addition, face information also changes with age. If the machine is unrecognizable, the user only needs to go to the system to update the face photo.
In order to improve the recognition rate, many application scenarios require the user to use dual authentication other than face recognition technology. Chen Jidong said that the cross-validation method will further increase the recognition rate, even if the twins are also "sent two." In fields such as finance, where the tolerance for false recognition rate is extremely low, even if the accuracy of the single identification factor is high, there will still be fish that are missing the net. Therefore, it is necessary to combine multi-factor comprehensive verification. At present, the accuracy of face recognition has far exceeded the naked eye, and a living body detection algorithm is used to determine whether the collected face information is a photo, a video, or the like. "Even if there is a very low probability of an account being misappropriated, Alipay will pay the full amount through the insurance company."
Doubt three: how to protect user privacy?
Some experts pointed out that compared with fingerprints and irises, facial features are biological features with weak privacy. For example, many people take selfies and are relatively open features. How to ensure the security of user data is particularly critical.
According to media reports, in a project called "Your face is big data", Russian photographer Yegor Cvetkov spent six weeks in St. Petersburg to take photographs of 100 subway passengers. Using face recognition tools to compare with the 55 million users on Russia’s largest social networking site VK (VKontakte), the profile of about 70 passengers was found.
How to prevent similar risks of privacy leakage? Xie Yinan, Vice President of Technology, said that despising the photos will desensitize the photos after they are captured, and only extract photo features, rather than the photos themselves. Even if these features are stolen during transmission, they cannot be restored. It is irreversible.
Chen Jidong said that Alipay has already protected face recognition technology from encryption and desensitization, and it can turn face information into an irreversible digital message that cannot be restored or compared.
Apple introduced that all its saved face information is protected in a safe compartment to ensure data security. At the same time, all processing is performed on the device and will not occur in the cloud to fully protect user privacy. The face ID is only unlocked when the user looks at the iPhone X and is specially designed to prevent face spoofing by the photo or mask.
Related industry standards need to be improved
Experts generally believe that the market potential of face recognition technology is huge. The technology requires high security, high accuracy, high availability, and high real-time performance. However, currently there is no industry standard for face recognition technology, and user privacy security is also urgently needed. Formulate and improve industry standards.
According to Shan Shiguang, a researcher at the Institute of Computing Technology of the Chinese Academy of Sciences, after many years of development, face recognition has indeed achieved breakthrough development in recent years and completed some "impossible tasks." However, user privacy is also worth paying attention to, namely, how the user's photos are transmitted and saved, and whether they have been saved or copied without permission. How related applications design face recognition systems to ensure that user data is not stolen is currently unclear. “Face recognition technology is gradually maturing and has more and more applications. Various standards for face recognition technology, including standards for protecting the privacy of citizens, should be introduced as soon as possible,†said Shan Shiguang.
According to Ms. Tian, ​​a technical personnel of the Huawei Group who is engaged in pattern recognition, face data is difficult to change. “For example, we cannot change our unique biometric code because one face data was stolen. Therefore, many technologies are now available. In the breakthrough bioassay test, such as 'open eyes' with 'open eyes' to confirm further."
Yang Fan said that face recognition is a long industrial chain. Protecting the privacy of users depends not only on the self-discipline of the company, but also the need to establish a unified standard for the entire industry under the guidance of the government and jointly build an industry dyke that protects the privacy of users.
Yan Shuicheng said that the basis for the wider use of face recognition is to further improve the recognition accuracy and safety, and that large amounts of data are essential for deep learning. In the future, more and more face recognition collection devices will accumulate large amounts of data. However, if these data become individual islands of data, face recognition technology cannot be improved. It is recommended to increase data sharing and openness. Guide efforts to promote technological development.