Face recognition technology is on fire again. Now, in a middle school simulation test in Beijing, students have to brush their faces. Since the iPhone X adopts Face ID technology, it has attracted attention, and people have fallen into the “sharp faceâ€. However, the brush face technology has just started, and its technology and experience are still immature. It is still difficult to develop on the road of science and technology, so we expect that the brush face technology will be highlighted on the road of technology.
“Sweeping face to check the provident fundâ€, “brushing face attendanceâ€, “brushing face security checkâ€... More and more daily behaviors are crowned with “brushing faceâ€, even in the public toilets of Beijing’s Ritan Park. Toilet paper also has to brush your face.
With the combination of face recognition technology and government affairs, you have to prove that you are yourself in the future. You may not need to cover more than a dozen chapters before and after, just brush your face.
Chinese people's strong interest in face recognition is also reflected in the financing of related companies: Face++, a face recognition cloud service platform, and Shangtang Technology, which focuses on building artificial intelligence visual engines, have received $100 million in Series C financing and 410 million. The B-round financing of the US dollar ranks among the "unicorns" in the field of artificial intelligence.
The tipping point for the commercial application of face recognition technology seems to have arrived. iPhone X may not be able to achieve itself, but it does make face recognition.
Three algorithms for face recognitionWhat is face recognition? If you want to identify the next definition of face, it is a technology that uses human biometrics to achieve individual differentiation, generally including image acquisition, feature location, identity confirmation and search. To put it simply, face recognition is the extraction of facial features from the image, such as bone features, eyebrow height, etc., by comparing the output.
Although the iPhone X's Face ID is exciting for entrepreneurs in the computer vision field, this face recognition is not the face recognition. Apple's Face ID does not use the ordinary camera, but uses infrared active recognition technology, so that it can achieve three-dimensional recognition and enhance the safety factor.
In a specific application scenario, face recognition can be roughly divided into three types: 1:1, 1:N, and N:N.
The 1:1 level of face recognition is achieved by the initial level of "Prove that you are you." It can be seen from the literal point that 1:1 is that the user uploads personal photos in advance and stores them in the system. Each time the verification is made, the offline photos are compared with the photo information stored in the system to determine whether you are not you.
For example, when we passed the security check at the station, the ticket inspector took your ID card and compared it with yourself to prove whether you are the person on the ID card. This scene is a 1:1 scene. Mobile phone unlocking, face payment, online ticket purchase, hospital registration, government Huimin project, as well as various securities account opening, telecom account opening, Internet finance account opening... This is the application scenario of 1:1 face recognition. Compared with other methods, the 1:1 recognition accuracy is high, and the calculation power is relatively low.
The 1:N face recognition algorithm is mainly used for face retrieval, "prove who you are." Unlike the one-to-one comparison of 1:1, 1:N requires a photo to be compared with a large number of photos in the system, and multiple comparison results are arranged according to the similarity. The results ranked first in the first place may not be accurate.
1:N face recognition algorithm is mainly used in the security field, such as for investigating suspects and finding lost children. In 2015, Yuntian Lifei, a startup that specializes in dynamic portrait recognition, has cooperated with the Shenzhen Longgang District Police to build a “deep-eye†system in local subway stations, railway stations, Chengzhong Village, and Shangchao. A few months after the launch, the police helped the police successfully smash two murders.
1:N face recognition applied in the security field, characterized by dynamic and non-cooperative. The so-called dynamic, that is, the system recognizes not the picture, but the video captured by the camera. Non-coordination means that the recognition object does not perceive the position of the camera and cooperates with the recognition work, and the recognition object is in a passive state. This improves the convenience of collection, but also means that your whereabouts have been completely exposed.
Compared with 1:1 recognition, the location, environment, light, acquisition angle and even glass reflection will affect the accuracy of 1:N recognition, so 1:N is relatively more challenging.
As for N:N face recognition, it is actually equivalent to performing multiple 1:N recognition at the same time for "proving who is who".
Face recognition technology is not perfectAs far as the basic research behind face recognition technology is concerned, China, Europe and the United States are almost at the same level. But when it comes to business applications, China is definitely ahead.
"Face recognition is not a very high-end technology. China's big companies pay more attention to immediate interests. Face recognition is the fastest and best way to realize artificial intelligence in their direction." Human Body Recognition Technology at Beijing University of Aeronautics and Astronautics Experts in the field seem to think that big companies like Google (microblogging) are not pursuing face recognition technology because they have longer-term plans.
Entrepreneurs in the field of artificial intelligence in China seem to be eager to get together.
From traditional Internet giants such as Alibaba, Baidu, Tencent, and Jingdong, to unicorns such as Face++ and Shangtang Technology, to the entrepreneurial team just entering the market, face recognition is not only a patent of a large company, but also a server. And the field of face recognition on the mobile side presents a trend of contending.
"The threshold for face recognition is already very low, but it is not easy to do it in real-life scenarios." Zhang Quanling, the founding partner of Zi Niu Fund, represents the views of many investors in the industry.
In the TV show "The Strongest Brain", Baidu's small-scale learning based on the depth of the robot basically shows the current level of development of face recognition technology - it can easily extract thousands of feature points on the human face and make depth through massive pictures. Learning training, exercise the pre-processing function of the face image, and can effectively match the face image in 1 to 2 seconds.
At the same time, the smallness also reveals some problems that face recognition needs to break through: poor lighting conditions, different angles, blurred information or distortion may cause recognition errors, in addition, face coverings, hats, beards, hairstyles, facelifts Or PS, etc. will also interfere with recognition.
This year's 3.15 nights exposed the technical loopholes in face recognition. The host only relied on a photo of the audience, and after technical processing, quickly generated a 3D face model identical to the audience. The host puts on the 3D face model of the audience, aligns with the mobile phone camera, and completes blinking, turning, smiling, etc. according to the instructions of the APP, successfully fooling the system and successfully completing the living body testing and certification.
In the interview, Qiu Xueyi, a visual analysis expert at the 360 ​​Artificial Intelligence Institute, said that the face authentication technology at this stage cannot be matured in all scenarios. Although the accuracy of face matching is very high, most face recognition. The system pays less attention to the detection of living organisms, and the algorithm used is relatively simple, and the cracking is not difficult.
The scene is the key to the entrepreneurial team breaking throughIn addition to the technology is not yet fully mature, technology-based startups often fall into a misunderstanding: only technical heroes. "Now most of the start-up face recognition companies are still at the level of 'My technology is better than others'." Preangel investment director Jiang Wei said that these companies have not yet thought about the commercialization scene.
Source source capital investment partner Zhang Hongjiang (microblogging) also said: "If the company only has algorithms, only a few cattle, no data or hard to get data, no application scenarios, such a company does not do much."
Of course, there are many companies that also emphasize data, but this is not an absolute threshold. "Assume that the existing technology has reached 92 points. If the new technology can reach 94 points, the difference in feelings for users is not so strong, and the meaning is not great." In Wang Jun, marketing director of Yuntian Lifei, the technology is always Iteratively updated, but when the technology can not be subversive, the value of technology will be greatly reduced.
Of course, this is not to deny the meaning of technology. It’s just that when the unicorn has appeared on the track, technology has become difficult to become a new company’s card.
"We are now paying attention to the face recognition company that has just started. The purely technical level is less, and more attention is paid to companies that can solve practical problems." Jiang Wei said that commercialization capability is the breakthrough point for emerging enterprises. AI startup teams with equal technical and business abilities are more likely to be capital-optimized.
Whether it is from the scene resources, data acquisition or capital strength, the giant has an unparalleled advantage of the entrepreneurial team. After the giant enters the game, how should the startup play?
“The foundation of a startup’s survival is the deep service and penetration of this industry.†An investor who did not want to be named said that the startup will not be easily replaced by the innovation of the giant. On the contrary, the massive investment and basic innovation of the giant will promote Opportunities for the new sector have emerged. For example, just after Apple promoted Apple's mobile phone and Google promoted Android system, the development of mobile Internet really broke out, and new opportunities such as Uber and Didi emerged. The emergence of Face ID is not a new opportunity for the face recognition startup team.
But if face recognition can mature applications, you still need to look at the scene. The above-mentioned investors said: "I am more optimistic about scenarios that are less demanding, such as consumption and entertainment, and are more cautious in areas such as financial payments where accuracy is extremely high."
This view coincides with Wang Jun, marketing director of Yuntian Lifei. “Former General Electric Chairman Jack Welch stated in the book The Essence of Business that the value of the product itself is more important. The world’s not the most sophisticated technology is the most commercially valuable, but I think The technology that is the easiest to replicate on a large scale is the most popular, so I am optimistic about the application of face recognition in areas where security requirements are not as high."
A Bridge rectifier is an Alternating Current (AC) to Direct Current (DC) converter that rectifies mains AC input to DC output. Bridge Rectifiers are widely used in power supplies that provide necessary DC voltage for the electronic components or devices. They can be constructed with four or more diodes or any other controlled solid-state switches.
Bridge rectifiers,Ultra Fast Switching Diodes,Fast Recovery Diode,Fast Switching Diodes,Diode Bridge Rectifiers
Changzhou Changyuan Electronic Co., Ltd. , https://www.cydiode.com