Recently, social media users have taken to posting the results of their submissions to how-old.net, a website Microsoft launched in April to demo its facial recognition technology, which guesses the age and sex of people in submitted photos - though not always accurately. In fact, the site's sometimes wildly inaccurate guesses have helped boost its popularity. Still, the market for facial recognition products has grown rapidly as companies have adopted the technology to improve security and customer service.

Alibaba Group Chairman Jack Ma Yun presents a new facial recognition feature of the Alibaba portal at the opening ceremony of the 2015 CeBIT technology trade fair in Hanover, Germany on March 15, 2015. Photo: CFP

When Beijing housewife Tong Ting submitted a photo of herself and her daughter on how-old.net earlier this month, the results were mixed.

The website, which guesses the age and sex of people in submitted photos, came close when it identified Tong's 13-month-old daughter as a 2-year-old girl. For the 30-year-old woman, however, the website's facial recognition model was well off the mark.

"It guessed I was only 5 years old," she told the Global Times Tuesday.

Tong was one of many people who have shared the website's guess with their friends on WeChat. Although much of the discussion focused on the poor accuracy, it wasn't always critical. It turns out a lot of people like it when someone, even a computer model, guesses that they are younger than they actually are.

However, facial recognition technology has uses beyond a carnival guessing game. Companies from a wide range of industries have adopted, or are considering adopting, the technology.

"In recent years, many of our clients have sought to use facial recognition technology to improve their businesses, particularly clients in finance," said Lü Sulan, who is in charge of the marketing department at Techshino Technology, a biometric company founded in 1997.

The facial recognition technology market has grown rapidly in China. Sales revenue for facial recognition products on the Chinese mainland grew from 670 million yuan ($107.89 million) in 2010 to 1.49 billion yuan in 2014, according to a report released by Beijing-based Bosi Data Research Center in March.

The technology

How-old.net received more than 210,000 submitted photos in the first few hours after it launched in April, according to media reports. Microsoft, which created the website to demo its facial recognition technology, has not disclosed how many photos the site has analyzed. The company did not anticipate how-old.net would become so popular.

"It was just designed as a demo - it is not for production use or to be compared with production services," Microsoft Research Asia said in a statement e-mailed to the Global Times on May 7. "It may or may not provide accurate guesses, but we didn't design it for that - we designed it to be a great example of how developers could build a fun app in a very short period of time."

Inaccuracies are to be expected, said Sun Jian, principal research manager from Microsoft Research Asia.

"The calculation method is based on a model with large number of samples. But it's still a tough problem to predict age accurately," Sun told the Global Times Friday.

"Microsoft owns more than 100 patents related to face-related technology. In the near future, the technology can be used for secure identification," said Chen Gang, principal development manager at Microsoft Research Asia. "In the long-term, the technology has the potential to be used to help robots and artificial intelligence better communicate with human beings."

The applications

In the financial industry, there are three primary uses for facial recognition technology, said Lü, whose company provides "multimodal biometric identification solutions."

First, it can be used to allow clients to open accounts remotely, which will save banks money on labor costs. To ensure the identity of client, the technology allows banks to verify images through three channels: a photo taken by the applicant, the image from users' ID cards and the image on file with the police system, Lü said.

Second, it can help bank tellers better identify users to enhance security.

Third, it can help bankers to better identify their VIPs to improve customer service. For instance, the system features a database of images of potential VIP clients, which it employs to recognize potentially valuable clients when they enter the bank.

"The competition banks are facing is so fierce that they want to be able to identify clients more accurately and provide good services," Lü told the Global Times Tuesday.

Bank of Jiangsu officially adopted facial recognition technology in its customer identification system in April.

Companies in other industries, such as e-commerce giant JD.com Inc, are also considering using this technology to improve their customer service.

"Currently, the company has not employed facial recognition technology," said Liu Shangkun, who is in charge of the recommendation and search engine department at JD. "But it could be a good supplement to help the company to better identify customers. Besides customers' shopping habits and characters, we need facial recognition technology to identify customers in order to provide differentiated services."

For image recognition technology, many companies are working for it, including Microsoft, Google and Baidu Inc.

Baidu's Minwa supercomputer scanned ImageNet, a database of more than 1 million photos, and taught itself how to sort them into roughly 1,000 different categories, according to Baidu.

The supercomputer can tell the difference between a loaf of bread and a meatloaf. In practice, humans can correctly identify all but about 5 percent of the ImageNet photos. Microsoft's software had a 4.94 percent error rate, and Google had a 4.8 percent error rate. Baidu had a 4.58 percent error rate.

Besides traditional application for a secure entry or time clock system, facial recognition technology can be used in other fields such as remote identification.

The market for facial recognition technology is ultimately decided by the population. China has an immense population, which makes it a potentially huge market, according to the report from Bosi Data Research Center.

"But customers should know that multimodal biometric identification is much safer than single biometric identification, especially when the technology is used in finance," Lü said. "We can't ensure the facial recognition technology can be 100 percent accurate, and it's safer if you can use other biometric identification together."