News

Development of Facial Recognition Technology

Development of Facial Recognition Technologyhe research on facial recognition system started in the 1960s. It was improved after the 1980s with the development of computer technology and optical imaging technology. It entered the primary application stage in the late 1990s actually, and it was dominated by the technical realization of the USA, Germany and Japan. The key to the success of facial recognition system lies in whether the key facial information is collected and whether the sophisticated core algorithm is possessed and the recognition results are made due to practical recognition rate and speed. “Facial recognition system” integrates artificial intelligence, machine recognition, machine learning, model theory, expert system, video image processing and other technologies. At the same time, it has to combine the median processing theory and realization and it is the latest application of biological characteristic recognition and its core technology realization shows the transformation from weak artificial intelligence to strong artificial intelligence.
The traditional facial recognition technology is mainly the visible light image based facial recognition. It is a familiar recognition way and has a R&D history of more than 30 years . However, this way has an insurmountable defect. In particular, when the environmental illumination changes, the recognition effects will decline sharply, and the actual system requirements cannot be met. The schemes which solve the illumination problem include 3D image facial recognition of visible light image and thermal imaging facial recognition. However, these two technologies are far from mature, and the recognition effects are dissatisfactory.
A quickly developing solution is the multiple light sources facial recognition technology based on active near infrared image. It can overcome the impacts of light change, and has achieved the outstanding recognition performance and its overall system performance in precision, stability and speed exceeds the 3D image facial recognition. Such technology rapidly develops in the past 2-3 years so that the facial recognition technology gradually develops towards practical.
Like other biological characteristics (fingerprint and iris) of human body, face is inherent. Its uniqueness and good characteristic of “not easily be copied” provide the necessary preconditions for the identity authentication. Compared with other types of biological recognition, the facial recognition has the following features:
1. Non-compulsory: the subject does not have to specially cooperate with the facial collection device. The facial image can be obtained in the unconscious state almost. Such sampling way is not “compulsory”;
2. Non-contact: the subject does not have to directly contact the device for obtaining the facial image;
3. Concurrency: in the actual application scenarios, multiple faces may be sorted, judged and recognized;
4. In addition, the visual characteristic is met: “recognize person by appearance” characteristic. Besides, it has such features as simple operation, visual result and good concealment.

Recognition Algorithm

Generally speaking, the facial recognition system includes image intake, facial positioning, image pre-processing, and facial recognition (identity confirmation or identity search). Generally, the system input is a piece of or a series of facial images with uncertain identity, and facial image of several known identities in the facial database or corresponding code. The output is a series of similarity scores, showing the identity of to-be-recognized face.
Generally, the facial recognition algorithms are classified as follows:
● Feature-based recognition algorithms.
● Appearance-based recognition algorithms.
● Template-based recognition algorithms.
● Recognition algorithms using neural network.
● Illumination based estimation model theory.
● Gamma grey level correction based illumination pre-processing method is raised. Besides, on the basis of illumination estimation model, the corresponding illumination compensation and illumination balance strategy are carried out.
● Optimized deformation statistics correction theory.
● Statistics deformation based correction theory, optimized facial posture.
● Enhanced iteration theory.
● Enhanced iteration theory is an effective expansion of DLFA facial detection algorithm.
● Original real-time characteristic recognition theory. Such theory lays emphasis on the median processing of facial real-time data, thus achieving the best matching effects between recognition rate and recognition efficiency.

Application Scenario

The facial recognition monitoring camera provides such functions as facial recognition, strong light inhibition, dynamic white balance, concealed sheltering, backlight compensation, and picture adjustment. It may be used for the surveillance video indoors and outdoors in the important parts of public activity and gathering places, such as parks, factories, stores, outdoor squares, conference centers, sports venues, schools, hospitals, residential areas, commercial streets and large supermarkets, and lobby entrance and exit, elevator, and other main channels of hotel (guesthouse), catering and recreation places and office buildings.
The identification/real-name system/real name verification/human-certificate verification/human-certificate combination may monitor the crowds at the airport, stadium, supermarket and other public places. For example, the monitoring system is installed at the airport to prevent the terrorist from boarding the airplane. In the bank ATM, if the card and password are stolen, cash will be withdrawn by others. At the same time, the facial recognition can avoid such circumstances. Through inquiring the target image data, whether basic information about key population exists in the database is searched. For example, the system is installed at the airport or station to arrest the fugitive.