For post-iPhone X models, Apple removed the Home button from the iPhone and instead activated Face ID for unlocking the device. Unfortunately, users had to resort to rather inconvenient passcodes during COVID-19 since they couldn’t easily unlock their devices with Face ID while they had their face mask on. But earlier this year, Apple made it possible for its iPhone users to unlock even with their masks on by updating the iOS 14.5. This begs the question: How does it work to tell a person apart only by one’s forehead and eyes?
Facial Recognition: its working & advantages
Facial recognition technology basically uses a method of storing facial features of the device user and comparing them to those of the person who wants to unlock the device. For the first part of this process, the technology identifies the face from the video or image and learns of the facial features such as eyes, nose, mouth, and facial lines from the face. Also at this stage, the data is iterated over and over as part of the AI’s deep learning process. The features of the learned face are stored as data in the form of vector values and then used to distinguish faces on a basis of how well the features learned by the AI match with those of the person who attempts to unlock the device. Finally the secret behind Apple’s technique to authenticate the user only by a portion of his or her face: if you subdivide the facial features by assigning more vector values around the eyes, you can distinguish the face even when you are wearing a mask.
Face recognition has a clear advantage compared to other forms of biometric technology like fingerprint, voice, and iris recognition - Facial data can be more easily collected and less likely to be damaged (wear and tear) by external factors.
China’s Lead in Facial Recognition System with its Overwhelming Amount of Data
One country that is dominant in the field of facial recognition is China. Chinese companies ranked both first and second in both categories of 1:1 verification and 1:N identification at the National Institute of Standards and Technology (NIST) Face Recognition Vendor Test (FRVT). To clarify a few terms here, 1:1 verification is to verify whether one picture of one person matches with another picture of the same person; 1:N identification is to identify multiple pictures of the same person from several pictures.
China’s lead in facial recognition is attributed to the nation’s lax regulations on personal information to the extent that the Chinese government allowed companies to access CCTVs installed at public places for datasets in them; The government’s full support of advancement of face recognition technology in the private sector; Last but not least, over 600 million security cameras nationwide in China and the government’s database of over 1 billion people’s faces from them. Now can you imagine how sophisticated the AI model would be since it was trained on the enormous amount of over 1 billion facial photos?
South Korea on a path to outpace China
Recently, an unusual and surprising event was witnessed in the China-led field of facial recognition. Kakao Enterprise, a subsidiary of Kakao, was ranked in top five in a total of four categories including being the best performer in the kiosk category at the National Institute of Standards and Technology (NIST) Face Recognition Vendor Test (FRVT). It was quite an event given that since 2018, Chinese companies have been sweeping from the 1st to 5th places in the competition. And these results show how quickly face recognition technology is advancing in Korea. Korean institutes may lag behind Chinese counterparts in terms of size of training data for now, but their engines are said to be on a par with Chinese ones.
Kakao Enterprise's next goal on AI CCTV (Closed Circuit Television). The company wants to identify a person by CCTV. To this end, it has opened unmanned convenience stores with AI CCTV within the company and updated the service from feedback from the visitors. With this technology in place, it is expected that unmanned stores can be operated only with CCTV without the expensive 3D cameras.
The more training data for the AI model, the more sophisticated the AIl becomes, they say. Crowdworks has over 400,000 skilled data labellers and has successfully responded to the needs of the industry and the whole AI ecosystem as an AI data platform company.
AI For Everyone, Crowdworks
Global AI Trend 11: The 5 Biggest AI Trends In 2023
Crowdworks unveils ‘comprehensive human QA system for AI data’