With the wind of face recognition technology in the domestic, a growing number of commercial applications emerged, particularly in the financial sector, before the first Internet Bank on the sea in your own APP integrates face recognition technology, safe and inclusive face recognition required to complete prior to the loan process, Jingdong Wallet provides unlock brush face ... ...
This reporter has learned, there at least more than more than 10 teams in the development and operation of face recognition, can be roughly divided into two camps, the first camp is part of the largest group within the research team and its related technologies for internal use only, the second camp to be independent of commercial technology companies, their foreign import and sale of technology.
And future, once Bank let go remote account, so, in financial field, people face recognition technology will ushered in big explosion of times, but this zhiqian, more bit industry people on newspaper reporter said, people face recognition at least also need solution three aspects of problem: how weigh customer experience and identification degrees Zhijian of relationship; how improve further run efficiency; how in people face recognition based Shang for living detection, and last a items vital.
Liveness detection, remote account or maximum constraints
In May, Microsoft launched How-old.net, by guessing age people face, and joy the whole circle of friends, but speculated that there is a small error, so many celebrities are "disabled". But behind this technology is, in fact, is face recognition.
Same application also includes a safe and inclusive in its link on the micro-circle of friends before a "measuring beauty deserves money" game, according to ping an insider, game theory is the use of face recognition, photo service through the cloud for data analysis, based on data identified photos of people in sex, age, degree of smile.
Of course, if applied to the remote account and other financial scenario, will not be so capricious.
At present, there are basically two types of face recognition development team, is rooted in the larger bodies, such as Tencent's excellent team, Ant-gold suit "paint shop" deep learning lab, Baidu, peace peace Innovation Lab technology; second, independent technology companies, such as Linkface, the wisdom eye, this kind of technology export-oriented team.
This which, most team is 2014 years zhihou appeared of, "actually people face recognition has research years, but why 2014 years zhihou suddenly became hot does, this is because people face recognition previously precision rate too low, below people eye, customer experience bad, 2014, Hong Kong Chinese University Professor Tang Xiaoou led of computer Visual research group development has a name for DeepID of depth learning model, the model get has 99.15% of recognition rate. It became a watershed moment, face recognition using machine begins to move beyond the human eye. "A face recognition technology research and development scientist told reporters. Nine launch four intelligence products small electrical
He said many research and development team, in accordance with its existing algorithm, enter large amounts of data to optimize, can achieve a high accuracy rate. But if applied to the financial sector, only accuracy is far from enough, also has a very high security, this involves the liveness detection, has proved that a living in front of the camera rather than a photo.
"At the moment, in academia, bioassay system is still at the exploratory stage. "The scientists even think, liveness detection, the maturity of the technology is the biggest constraint remote account opening or not.
Micro-banks of liveness detection methods is that the user needs to the camera read out a string of random numbers and Christmas Whitney APP liveness detection methods is to ask the user to follow the instructions and make decisions such as shaking his head and mouth wide open, nods and other actions.
But quite a few users, safe and inclusive face recognition is too difficult, and appear as you do this yourself retarded. Reporter experience shows that through face recognition has two requirements: first, take off your glasses and, secondly, orders jumped out very quickly, and requires the user to be so slow to respond. Reporter stuck on the first step: after you take off your glasses before a vast expanse.
This involves the liveness detection, identification and customer experience problem, both like a seesaw, "with the commercial application of face recognition, secure and convenient, in fact, customer demand for more convenient, we have to do is reach a relative safety, accepted by public safety, increase the cost of fraud. "Linkface co-founder, said to the reporter. Under the premise of not affecting the customer experience in order to increase security, face recognition also began to combine with other biometric technologies.
On August 28, the HKUST, fly, Linkface, China UnionPay in 2015, the third session of the world trend of IC Card Forum mobile payment Conference held a news conference to launch "voice recognition + face recognition", it is understood that the Merchants Bank will be the first application of this technology in the P2P transfers: when users transfer, you need to read 8 random numbers into the phone.
Above Linkface co-founder, introduces the technology involved "voice recognition + die + face," "our previous results showed that this confirms the precise rate of 99.9%. "She said.
Speed is about customer experience
In fact, this reporter interviewed a number of research and development institutions to face recognition, they both agreed that its face recognition technology to identify more than 99%.
Peace General Hui brand Director Cheng Rui told first financial daily reporter, currently, peace General Hui by using of people face recognition technology precision rate up 99.5%; Ant gold clothing Chief Strategy Officer Chen long told newspaper reporter, its in internal stage of test inside, people face recognition of accurate rate reached has 99.8%; Tencent said, its people face recognition LFW data set accurate rate over has 99.5%. In actual practice the product social networking accuracy rate of up to 99% on the image, for ID photos accuracy than in the 99.9% above Linkface co-founder told the newspaper, its research and development of face recognition accuracy rate of 99.5% per cent.
According to one people face recognition industry people introduced, in people face recognition process in the, roughly will experience four a steps: first step: people face detection, positioning people face in where; second step: people face key points positioning, like in people face Shang added a lipstick or eye shadow; third step: recognition is who; fourth step: people face property, recognition people face of mood, have laugh, have wearing glasses, hair of color,.
According to the specific application of face recognition, these four steps can be arbitrary choice, such as applications in the financial field, almost all include these four points. And if just one face photo album, may need the first three steps.
In addition to in vivo testing, customer experience, and face recognition technology is still a problem to be solved is the speed, "at present, the application deep learning training model for face recognition are usually larger, which we need at the algorithm level adjustment and optimization in engineering, further improve the speed. "The Linkface co-founder said.
Now, with the depth of penetration of Internet technology to every industry, face recognition market will be more promising, forecasts show that within the next three years, our face recognition industry's annual sales are expected to have tens of billions, over the next decade is expected to be formed had annual sales of billions of dollars in market size.
Newspaper reporter survey found, currently provides people face recognition technology of technology company main has two species charges way, a is charged one-time technology fee, a is by times charges, like, industry a home company of quotes is identity certification 4 Yuan/times, actually, people in a years among to completed hundreds of times identity certification, as purchase train votes ticket, in all financial institutions account, and login, and paid, and transfers,, according to 4 Yuan/times of charges calculation, future billion sales scale can period.
Actually, increasingly more of application will was development out, in past, a girl Lovelorn has, she may need turned out past and Qian male friends of photos, a Zhang a Zhang to wipe off Qian male friends of face, now, she just to in people face album in the, through people face recognition in the of people face poly class technology, a key brush off all Qian male friends of face.
No comments:
Post a Comment