Many software in the identity security authentication, face recognition is an important link. But do you really think that face recognition is just to shoot the “face”? Recently, the digital blogger revealed in the social platform: face recognition to collect the area is not only the screen display of the head, but the entire scope of the camera coverage. And the photos are uploaded to the backstage, where human reviewers can see them.
Soon after, the hashtag # Facial Recognition Must Wear Clothes # was trending. Many exclaim that if that’s the case, they’re going to another planet. So can background reviewers really see more of us than our faces on the phone’s screen?
In this regard, Professor Qiu Bo, director of the Department of Electronic Information Engineering, Hebei University of Technology, affirmed that “the image taken by face recognition must be all the areas covered by the camera field of vision, rather than just the part within the frame we see on the phone. This is a problem of basic common sense”.
“From a technical point of view, the current face recognition technology does not require the storage of the original image.” According to Qiu Bo, the intelligent camera can extract the features of face images in real time and encode the faces, thus generating a face feature vector for transmission, storage and comparison operations. That is to say, the terminal machine, face have already turned into a string of Numbers, they can be said of the distance between the eyes, the size of the position of the corners of the mouth, face, skin roughness, etc., and exactly what is according to the different face feature calculation methods will have change, after these characteristics calculation every face turned into a “password” – feature vector.
“When the face is converted to vector value, the machine’s face recognition process is similar to looking for a specific password in the password book. It just needs to compare the numbers.” Qiu Bo said, it can be said that from a technical point of view, face recognition can be done at the vector level.
For everyone concerned about the manual audit into a large “social death” scene, Qiu Bo said there is no need to worry too much. For large companies, the workload of face recognition needs to be very huge every day, and almost all of this work is done by machines. And, now provide face recognition technology of the head of enterprises and large enterprises are using privacy computing technology, they will only provide customers with desensitization feature code for comparison, will not be sent to the background image. It is also uneconomical for companies to send raw images to the background, simply because of the cost of storage and transmission. As a result, situations in which a human in the background can see images are “highly likely to be out of compliance.”
“However, there are also companies that store photos of their users for strategic reasons, in order to diversify and improve their algorithms. “By using machine learning to optimize algorithms for face recognition, we can develop algorithms that are safer, more concise and more accurate.” Qiu Bo said, from a technical point of view, the more information collected from the original image, the more trouble to face recognition. For example, the image background is very complex, there is a poster of a star, so the computer must first do face positioning, and even face recognition comparison on the poster, increasing the additional difficulty and calculation. So for the general company, will use the means such as setting a face frame to narrow the need to upload the processing of the camera area.
Qiu Bo stressed that although the current regulations have a restrictive effect on enterprises, there is no guarantee that no one breaks the rules in the whole process. So the best way for the public to protect themselves is to understand the risks and try to avoid them.