In the process of working on the next generation of Pixel smartphones, Google is collecting data to improve facial recognition. Employees of the corporation go to public places in large American cities and offer some people to take selfies on their phone. Those who agree are given a five-dollar discount coupon in return for purchases from Amazon or Starbucks.
Photos of the face are taken with an unknown smartphone, hidden in a weighty case that completely hides the design of the device. People who took part in a kind of action say that they were asked to take several pictures of their faces from different angles. Zdnet told one such story: an editor's friend, George, was doing relaxation practices in a park in New York when Google employees approached him. He watched passers-by and noticed a couple of people who looked like social poll volunteers or Scientologists. They addressed him with the words “Hi, we are collecting data to improve the facial recognition function in future Google devices”.
As a real inquisitive engineer, George agreed. Before the actual shooting process, the data collectors took his written consent to the processing of the data. According to him, the very fact of transferring his biometrics to the IT giant does not bother him, because Google already knows a lot of details about him. Facial recognition for user authentication requires fast and accurate operation of sophisticated artificial intelligence algorithms. The underlying neural networks must initially be "trained" on a large amount of input data. This seems to be what Google is doing, but it's not clear what the scale of the campaign is. For effective training of recognition systems, even fairly simple images, usually tens of thousands of images are required.
Neural networks work in a similar way to the human brain. By learning from vast experience, they build algorithms that are most optimal for performing the required task.
Previously, some companies have already been caught using large databases of human faces to improve their neural networks. Despite the fact that most often it was done absolutely legally, the American public perceived this fact in a rather negative way. Face authentication systems have become the most popular way to unlock smartphones in recent years. There are two main approaches, differing in the complexity of the implementation and the level of security provided. The first and most common is simple face recognition using a conventional front camera. It is cheap, moderately demanding on the resources of the smartphone's chipset, but at the same time it can be fooled by a correctly taken photo. The second involves building a height map of the user's face, which is illuminated by an infrared LED. A special camera measures the distance up to several dozen key points and, in fact, “recognizes” not the face, but its three-dimensional model. The reliability of this option allows Apple customers to use it to authorize financial transactions on smartphones.
What kind of face recognition system will be installed in the next generation of Google Pixel is still unknown. On the one hand, judging by some leaks, biometric user authentication "by the face" will be as reliable as the iPhone. This follows from the analysis of the beta version of Android Q, which is expected to be presented for the first time along with a new smartphone from Google. In its settings, a number of items were found that are characteristic of full-fledged biometric authentication, similar to that in Apple devices. On the other hand, the well-known Pixel 4 design concepts do not include any additional sensors on the front besides the regular camera.
Vasily Parfenov