The era of facial recognition is here #Oriental Outlook#-Sino-US


Path: Sino-US >> Cover Story>> >>
The era of facial recognition is here #Oriental Outlook#
In many European and American science fiction movies, there are scenes of rich people having to authenticate their identity with fingerprint, face and iris, so that they can enter the room with the highest level of security and confidentiality. 
 
On September 1, at the KFC's concept restaurant KPRO, customers can look up into a camera to order and pay for their meals. The same month, iPhone X with Face ID went on sale. 
 
Services like people withdrawing money and taking packages by facial recognition are likely to follow soon. 
 
Facial recognition is just a leading example of the biometric identification technologies that also include fingerprint, iris, voice and so on.
 
In the 46th issue of 2017, Oriental Outlook magazine under the Xinhua News Agency ran a cover story on the development and application of China’s biometric recognition techniques, as well as the concerns over the identification technologies.
 
Below is an excerpt of the article.
 
The country’s booming Internet industry has raised high requirements for information security in recent years, and that has demands further improvement in the identification technologies.
 
Around 2014, authorities issued regulations on using real names to use mobile phones and logging onto the Internet, which created the need to develop identification technologies.
 
China’s biometric identification technologies started about decades later than the Western countries, but the country has made great achievements in the core algorithms, data accumulation and model building.
 
China’s iFlytek Co Ltd is a worldwide leader in voice recognition, attracting many foreign companies to cooperate with it, and the EyeSmart Technology has advantages in iris recognition, becoming the technical service provider to the India’s government-dominated national biometric identification system.
 
According to a report issued by the Forward Industry Research Institute, China’s biometric identification market grew 60 percent annually between 2010 and 2014, and by 2020, the market will reach 30 billion yuan.
 
The biometric identification technologies have now been used widely from security to telecommunications, payment and retail. Different identification technologies have also been combined for more accurate recognition.
 
But all of the identification technologies have certain problems. Facial recognition is not so accurate in a complex situation. Iris recognition requires the person to be at a short distance, and the voice recognition has screen off noises.
 
To solve the problems, China and Western countries are at the same starting line. These problems could not be solved overnight, which rely on the computer and Internet technology development.
 
Anyway, it is predicted that China will see a rapid development and application of the identification technologies in a broad and vibrant market of 1.3 billion people.
 
From ideal to reality
 
In September, Alipay first put facial recognition technology in commercial use at KFC's concept restaurant KPRO in Hangzhou. Users can use facial recognition function on the Alipay app to pay for their bills within 10 seconds.
 
Apple Inc pushed ahead the Face ID function on its iPhone X mobile phones, not using its former fingerprint identification technology. The function will scan the users’ faces to identify owners and unlock. 
 
“Except Apple phones, many home-grown phone producers have also started using facial recognition technology in their products, such as Xiaomi, VIVO and Huawei,” said Xie Yinan, deputy president of Beijing-based tech firm Megvii Technology.
 
Some banks including China Merchants Bank began to use the facial recognition technology for customers to withdraw money from ATM machines. The amount was once limited to 3,000 yuan through the function. 
 
Some self-service supermarkets also allow people to take goods through facial recognition. Universities started to use iris recognition to check the attendance. 
 
An application era of the biometric identification technologies has really come, said Shan Shiguang, a researcher from the Institute of Computing Technology Chinese Academy of Sciences, who has worked on facial recognition since 1997. 
 
“In the past several years, the deep learning technology has pushed the biometric identification technologies onto a phase that a lot of products have come into real life,” said Shan.
 
“Chinese customers have been ‘educated’ in the Internet-plus era, and they would like to accept and try some new things,” Shan said. 
 
He said in the recent worldwide trend of artificial intelligence application, Chinese companies could grab many opportunities in the biometric identification application, and they are already leading in some fields.
 
Megvii Technology that was found in 2012 focusing on facial recognition has worked with the public security authorities to set up the sky eye system to use big data and facial recognition technology to catch more than 3,000 suspects.
 
Xie said that in such application the facial recognition technology must secure an accurate rate above 95 percent, and give responses in real time, which makes high demand on the algorithm and computing power.
 
To open bank accounts, customers should take their identification materials to a bank where workers will check their authentication. Following the development of the Internet finance, customers could open bank accounts online.
 
Xie said that customers needed to upload their identification materials for comparison at first, and now they could be identified through facial recognition and withdraw money through facial recognition.
 
Gong Yazhuo, general manager of Shanghai-based Irisian, a company focusing on iris recognition, said that his company offered banks with a plan combining the iris and facial identification technologies.
 
“Facial recognition could be used for long-range identification, and iris recognition could be used for close range identification,” Gong said.
 
Ma Li, CEO of Beijing-based iris recognition company IrisKing, said that the fingerprint technology is low-cost, efficient and convenient, and it can be used together with the facial recognition and iris recognition.
 
He said to use mobile phone to reply to messages in office, it is more convenient to use fingerprint measures rather than facial and iris recognition, but to ink a big deal, using the last two to secure accuracy is better.
 
In the future, the identification technologies will be used in different situations, and the facial recognition will be the mainstream, with fingerprint and iris used as alternatives, Ma said.
 
Data complexity 
 
Talking about the future of the biometric identification, Xie visualized that when people walk into a shopping mall, sellers could immediately identify customers’ habits and preferences through a recognition system, and customers pay for their goods through the system too.
 
“It depends on the accuracy of the biometric identification, the first step. It will surely optimize the whole shopping process, improving customer experience,” Xie said.
 
Put simply, the biometric identification technology is actually data comparison by a machine, determining the matching to finish the recognition task, Shan said.
 
The core of the machine is a function produced through multiple artificial neural networks, Shan said, and the biometric identification technology relies on the improvement of the deep learning technology.
 
Shan said that by inputting a large amount of photos, including of seniors, children, the black and the white, with different scenes and light situations, the function can become more accurate. 
 
Gong Yazhuo said that compared to companies with strong algorithm, the social network companies like Facebook and Tencent Youtu Lab have big data and it would be easier for them to succeed in the facial recognition field.
 
Tencent Youtu has developed its algorithm and unique technologies for quick identification based on the data from large amount users of Tencent’s WeChat and QQ, said Huang Feiyue, director of Tencent Youtu.
 
Tencent Youtu has achieved a record face recognition rate of 99.80 percent in LFW (Labeled Faces in the Wild) Database in 2017, and gained 83.29 percent accuracy rate in the MegaFace identification accuracy with 1 million distractors.
 
“We take part in the competitions, aiming to test and verify our technologies, and it will be meaningful if the competitions could reflect the advantages and disadvantages of the technologies and boost the development,” Huang said.
 
However, an industry insider said that in the real scenes, the accuracy rate will only be about 70 percent, and people wearing glasses or changing their hair styles might not be recognizable to the machine. 
 
Min Kerui, chief scientist of Beijing Orion Star Technology Co Ltd focusing on facial recognition, said that the accuracy rates were created with specific data in competitions, and the real scenes would be more complex.
 
In the competitions, the photos used were fixed, but in reality, people have different gestures, different expressions, and there are different light situations and obstacles, which raise the difficulty level for the machine, said Min.
 
“Situations are important in voice recognition. For example, amplifying the voice and lowering the noise all involves computing ability, a big database and algorithm, which influence the accuracy rate of recognition,” Min said.
 
He said the database of facial recognition was stable in the competitions with about thousands to hundreds of thousands of photos, but if there were millions or billions of photos, the machine might do the identification much slowly, with low accuracy rate.
 
Xie Yinan said that the accuracy rate, response time and meeting extra conditions are key factors to test if the facial recognition technology is practical or not.
 
Different scenes need different technologies and there is “no common solution,” he said. “Quick recognition is very important, such as using facial recognition to open a door.”
 
Safety concern
 
China has developed home-grown biometric identification technologies and products, but there are some problems and challenges, said Sun Zhenan, secretary general of the biometric identification technological innovation alliance.
 
There are four aspects of the safety issue, including the front end prosthesis attack, computing template theft, hacker attack during transmission and data leakage, Sun said.
 
The prosthesis attack is related to using photos, videos and models to create false materials to pass the recognition, which requires the technology to carry on the living body certification.
 
Tencent Youtu’s plan is to combine the lip language in the facial recognition to identify a living body. Alipay requires customers to perform a random act, such as blinking or shaking head.
 
“The verification on living body could also be carried out by improving the sensors to check the live skin and detect the heartbeat. The photos and the plastic 3D models would not pass such a test,” Shan Shiguang said.
 
“If the template was stolen, the system could automatically recognize the stolen template, cancel it and create a new template for the same biometric feature,” said Ma Li, CEO of IrisKing.
 
Ma said that the current biometric identification technology uses cryptology method to encrypt data and coding, which makes each arithmetic irreversible. “If the encoding of pictorial data was stolen, it will require a supercomputer to work 1,000 years to decode,” he said.
 
Hacker attacks and data leakage have become the most serious problems. For example, India has set up a national biometric identification system covering 1.2 billion people through iris recognition technology, but Indian government recently found that more than 210 official websites have released the details of citizen’s identification information, including fingerprint and iris data.
 
Sun said that the safety problem could not be eradicated at present, and all measures for anti-counterfeiting and loss prevention were not mature enough to solve the problem. “It depends on the improvement in computing and Internet technologies.”

Related Stories
Share this page
Touched Sympathetic Bored Angry Amused Sad Happy No comment

The era of facial recognition is here #Oriental Outlook#Introduction of Property tax expected to be accelerated with long-term mechanism for China’s property market lacking8 terms that matter in China-U.S. relations in 2017Capital's air quality sharply improvedNostalgia spells successChinese firm to recycle 'panda poo' into tissue paperTencent works with publishing house to develop augmented reality textbooksBusinesses' use of livestreaming illuminates privacy concernsKiller of Chinese graduate student in Japan sentenced to 20 years’ imprisonmentVoluntary rescue efforts in China dampened by unexpected death
< Prev Next >