Time of Face Swiping/Face recognition
Do you remember these plots in the movie? In "Transformers 2", when the young actor and his friends passed through the checkpoint, although they successfully cheated the soldiers on duty, they were discovered by the military's face recognition technology. In the 2014, remake of "Mechanics", the first time the Mechs appeared to the public; they scanned all the faces in the crowd, and compared the acquired faces in the wanted criminal database. Yes, it instantly found a wanted man who had escaped for many years among the crowd bystanders and subdued it. There are many other films that in the confidential departments of the United States, all kinds of biometrics must be scanned when entering the door, from fingerprints and irises in early movies to human faces.
What exactly is face recognition?
Face recognition is a subdivision term of visual pattern recognition, and it is probably the most difficult problem to solve. In fact, our people are recognizing visual patterns at all times. We obtain visual information through our eyes. The information is recognized as meaningful concepts by the brain so that we knew whether it was a glass, a book, or something else before us.
We also do face recognition all the time. We meet countless people in our daily life, recognize those acquaintances, greet and communicated with them, and ignore other strangers. Even avoid those who we owe money but not yet paid back.
However, this seemingly simple task is not so easy for the machine to achieve.
To the computer, image information, no matter it is a static picture or a frame in a dynamic video, is a matrix of many pixels. For example, a 1080p digital image is a matrix composed of 1980* 1080 pixels. If each pixel is in 8-bit rgb format, it is 3 numbers ranging from 0 to 255.
In these data, the machine needs to find out what kind of concept a certain part of the data represents: which part of the data is a glass, which part is a book, and which part is a human face, which is a rough classification problem in visual pattern recognition.
For face recognition, it is to distinguish whom the face belongs to in the part of the data that all machines think is a face. This is a subdivision classification problem.
Face detection and face recognition
There are several steps to complete the face recognition. First, the computer needs to find the position of the face in the image or video. This part of the work is generally called face detection. As mentioned earlier, this is a rough classification. When it comes to face recognition, it is actually a secondary-category classification. The computer only needs to determine whether the target image is a face. However, since the size and position of the face cannot be determined in advance, the computer needs to scan the whole image with each possible face size, and determine whether the image captured by the sub-window is a face one by one. In each scanning process, the sub-window may move by several pixels.
Therefore you can roughly imagine how many times the computer needs to make a secondary-category judgment to detect a picture.
The face detection obtains the position and size of the face from a picture, and sends the part of the image to subsequent steps, which include positioning of face parts, alignment and normalization of the face image, and selection of the quality of the face image, feature extraction and feature comparison. After all these steps are completed, the identity of the face can be known.
Of course, we can also use the face detection function alone to complete certain applications. For example, most current cameras and mobile phone cameras have a face detection function, which can automatically obtain the position of the face, so as to automatically adjust and optimize the picture. Even make some preliminary judgments on the face, such as gender, age, and even face score.
1v1 face verification and 1vN face search
The protagonist passed several checkpoints of identity verification and successfully entered a confidential department through various means. This is a plot that often appears in movies. These checkpoints of identity verification often include face recognition. In this application, users often need to provide their own identity.
For example, when giving a door card, the computer obtains and maintains a face pattern of the door card owner in system. When a person using a door card, the system compares it with the current face image of the person that are using the door card to confirm whether the person who are using the door card is the owner or not, so you can avoid the people who picked up your card to easily get into the company.
This is a 1v1 identity verification. The computer performs a comparison between the current face and the stock face, which is an auxiliary to other verification methods, thereby improving the reliability of identity verification. This kind of application has been widely used at present, such as accessing to sensitive facilities, remote account opening in the Internet finance field, and identity verification for large amount withdrawal.
The bridge section in "Mechanics" mentioned at the beginning of the article is a 1vN face search. The mechanical police can search online for a face database that stores all wanted criminal data. Every time he meets a person, he will first obtain the face information of that person, and use the obtained information to compare the wanted criminal database one by one. If the match is found to be high enough, they will be arrested on the spot. Each time of face recognition, the computer should make n times of face comparisons; n is the number of face patterns in the database to be recognized.
If the computer is required to recognize a person's identity only by the face, this is actually a type of 1vN face search. The target face database is an "acquaintance library" composed of n faces, with the increase of n, the difficulty of accurate recognition will increase, and the calculation time required for recognition will increase. We can image how many faces can an ordinary person accurately recognize and it's probably dozens times of the figure that 1vN can achieve.
The current best face recognition technology has actually exceeded this level. For example, the top domestic face recognition companies generally have a screen wall to demonstrate the personnel activities captured by the company's cameras and accurately identify their identities, and the company generally maintains a face database on the number of a hundred level people. However, if N continues to increase, reaching thousands of people and thousands of people, then finding the only matching face in real time becomes a sci-fi requirement. In larger face library applications, the real-time performance is generally reduced. It is only required to find the suspicious faces, normally consider several top similar ones, so as to narrow the scope of manual retrieval.
Application of face recognition technology in life
Many mobile phones are still using fingerprint or password to unlock, but some manufacturers have already begun to use the unlocking scheme based on face recognition technology developed. When the camera is turned on, the camera is aimed at own face, and it can complete the instantaneous Identity verification, then the mobile phone will become safer, even if someone else picks it up, it will be quite difficult to unlock it. With such a mobile phone, would you still worry about the insecure of funds in the mobile phone?
Many students still need to press the fingerprint when they take the exam for identity verification. Many times, because of tension and sweating on their fingers, they have to press several times, which is low efficiency and also brings psychological pressure to the candidates. But if using the face recognition technology, there’s no worry about it. The high accurate recognition rate can effectively overcome such risks. Let us think about it. If face recognition technology is applied to police cases, will it greatly improve the efficiency of handling cases, or will it make criminals have nowhere to escape.
Many companies have access control systems. Traditional access control systems often have problems and are not efficient to use. For those large enterprises with tens of thousands of people, the management of workers is a big problem. Who comes to work and who does not come, counting them one by one will be a very complicated job. But with the face-recognition technology, everything is from worrying, and employees have already completed attendance records at the same time they come and go.
Senad office entrance permission system