02 Apr ,2020
According to media reports, there are sellers selling hundreds of thousands of face pictures wearing masks at the price of two corners. Some of these photos are facial photos collected by "web crawlers" from the Internet, and others are facial photos taken when people check in at work or enter or leave the community. Buyers buy these photos, which are mainly used to train the accuracy of the algorithm, and may also be used to apply for a credit loan or even register a company.
Facial data breach
Under the epidemic, some technology companies have developed face recognition technology for wearing masks, which has brought convenience to people and reduced the risk of virus infection caused by removing masks. But for those who are already worried that face recognition technology will cause data leaks, this makes them even more worried: even wearing a mask will not stop facial data leakage. The above media reports have made their anxiety a reality, and at the same time, they have revealed that the leakage of facial data exists in every aspect of this technology from development to application.
Face recognition technology is based on the extraction of a large number of feature points on a human face. For a person wearing a mask, they can only extract feature points in their eyes, eyebrows and other areas, and cannot perform a full face scan. At present, the main implementation method of face recognition technology with masks is to strengthen the recognition of key areas such as eyes and eyebrows through model training. Some also reconstruct face images with masks into faces without masks through image reconstruction networks. Image, and then realize face recognition through comparison.
No matter which implementation method is used, the accumulation of human face data is required, and the more massive the data, the better the accuracy can often be improved. Where does this massive data come from? The above report provided a clue: a large amount of face data was illegally sold and used to train the accuracy of the algorithm. Although it is not possible to match buyers with companies that develop new technologies, companies that launch new technologies should announce the source of the face data required for training.
Technology is neutral
If the basic data used for research and development is unclear, it is difficult to say whether the technology is an angel or a devil. In addition, it can not be ruled out that some criminals are also developing this technology. They purchase data and train the accuracy of the algorithm in order to more easily crack the face password.
These basic data may be leaked in various links such as storage, transmission, and application. This is due to both the loopholes in the technology itself and the reasons for some theft. Some even guard against theft. It is not uncommon for people responsible for data management to sell data. Although specific cases may be found for a single data breach case, a large number of leaked data are combined, and the reasons behind them are complex, and it is difficult to say exactly how it was leaked.
The immutability of the face makes many people conservative about face recognition technology. The "Face Recognition Landing Scene Observation Report (2019)" released at the end of last year shows that in the situation of frequent leakage of personal information, more than 70% of people have doubts about the security capabilities of network operators and are worried about face data. Give way. Last year, a university professor was forced to use face recognition to enter the park by the Hangzhou Wildlife Park, and the latter was brought to court in anger.
How to collect the basic data required for face recognition technology? Where is the boundary of technology application, and how to grasp legitimacy and legitimacy? There is no clear statement yet. Relevant departments also have certain difficulties in supervision, and more rely on industry self-discipline, which makes the data leakage very uncertain.
The last line of privacy
Although many technology companies claim that face recognition technology is ready, and face recognition technology with masks has also been applied to some companies, how to prevent face data leakage is rarely mentioned. Important matters related to personal privacy and safety. Before social regulations are not ready, the scope of applications should be strictly restricted and user consent should be obtained. Build a high defense line and apply it no later. After all, the face may be the last line of privacy for people. If this password is cracked, the consequences will be unimaginable.