Home  Semiconductor Information

Aug 15 2018

What Is Facial Recognition Technology?


Facial recognition technology (FRT) is based on the facial features of a person, using the input facial image or video for face recognition (identity confirmation or the Search for identity). First, determine whether or not there is a human face, and if the answer is yes, then detect the size and positions of the major facial organs of each face. Based on this information, further extract the features from faces and compare these with known faces to identify each of them. Broadly-defined face recognition actually includes a series of related technologies for constructing the facial recognition system, including facial image acquisition, face localization, facial recognition preprocessing and identity confirmation & search; facial recognition in a narrow sense refers to a technology or system for people search or identification through human faces. 

The development history of facial recognition technology 

Embryonic stage (1964~1990)  


At the very beginning of facial recognition, it is only studied as a general pattern recognition problem, and the main technical approach is based on facial geometric structural features. This is mainly reflected in the study of profile and a great deal of research on the extraction and analysis of the structural features of facial profile. Artificial neural network has also been used by researchers for face recognition.

Growth stage (1991~1997)


Although this period is relatively short, it is the climax of facial recognition research, having yielded fruitful results: not only gave birth to several representative face recognition algorithms been born, but the US military also organized famous FERET tests; in addition, a number of commercially available face recognition systems have emerged, such as the most famous FaceIt system (now Identix). The "feature face" method proposed by Turk and Pentland of MIT Media Lab is undoubtedly the most famous human face detection method in this period.

Expansion And Maturity (from 1998 until now)


In recent years, due to the development of computer technology, the research of facial recognition has attracted more and more attention. Among many research directions, the most popular one is about the frontal face model, which can be divided into three stages:

The first stage is to study the facial features needed for facial recognition.

The second stage is the human-computer interaction recognition.

The third stage is the automatic recognition.

The face recognition is mainly focused on two-dimensional images which mainly uses 80 nodes distributed from low to high on human face to measure the distance between eyes, cheekbones and so on. Face recognition algorithms mainly include:

1.Template Matching: Templates are divided into two-dimensional templates and three-dimensional templates. The core idea is to use the law of human face features to establish a stereoscopic and adjustable model frame, and use the model frame to locate and adjust specific facial parts after getting the human face position so as to avoid recognition errors caused by different observation angles, unclear facial presentation and expression change during the process.

2. Singular Value Feature: The singular value feature of the face image matrix reflects the basic properties of the image and can be used for classification and recognition.

3. Subspace analysis: It is widely used in face feature extraction and has become one of the mainstream methods of face recognition due to its strongly describable feature, low computational cost, and being easier to implement or being separated.

4. Locality Preserving Projections (LPP) is a new subspace analysis method. It is a linear approximation of the nonlinear method-Laplacian Eigen map, which solves the problem that traditional linear methods such as PCA are difficult to maintain the nonlinear manifold of the original data, and the shortcoming that it is difficult to obtain low-dimensional projection of new samples by using nonlinear method.

3D face recognition technology

The 3D face recognition can greatly improve the recognition accuracy. The real 3D face recognition is a technique using depth images, and we have made great progress since the early 1990s. The three-dimensional face recognition methods are:

1. The method based on image feature: That is an algorithm that acquire attitude estimation from the three-dimensional structure. First, matching the size, contour and the three-dimensional direction of the face; then, when the posture is fixed, allotting different facial features on the model. (These features are artificially identified by researchers.)

2. The method based on variable Parameter of the model: Combine the 3D deformation of the universal face model with the matrix iteration based on the distance mapping to recover the head posture and the 3D face. Those parameters will change along with the changing model formation, and the process is repeated until the minimum scale meets the requirements. The biggest difference between the model-based variable parameter method and the image-feature-based method is that the latter needs to re-search the coordinates of the feature points after each change of the face pose, while the former only needs to adjust the parameters of the 3D deformation model.

The advantages and disadvantages of face recognition technology and its hidden risks



The so-called naturalness refers to distinguishing and confirming identity by observing and comparing faces.

2.not easy to be detected 

The recognized face image information can be actively acquired without being perceived by the tested individual. Face recognition is to obtain face image information by using visible light; and different from fingerprint recognition or iris recognition, it doesn’t need to use an electronic pressure sensor to collect fingerprints, or use infrared light to collect Iris images, these special collection methods are easy to be detected. 

3.non-contact mode

Compared to other biometric technologies, face recognition is non-contact and users do not need to be in direct contact with the device.


In practical application scenarios, face recognition technology can sort, judge and identify multiple faces at the same time.


1.There is similarity in human faces.

The differences between different individuals are small. The structures of all faces are similar. In addition, the masking of makeup and the natural similarity of twins increase the difficulty of recognition.

2. Variable face

The shape of the face is very unstable, and people can produce many expressions through facial changes. At different viewing angles, the visual images of the faces are also very different. The face recognition is also subject to other factors, like illumination conditions (such as day and night, indoor and outdoor, etc.), face coverings (such as masks, sunglasses, hair, beards, etc.) and age. 

3.Increasing market demands require updated technique.

The probability that people looks alike is increasing as the number of people who is under identified surges, so the original face recognition application can not meet the practical demands.


Privacy Rights Issue

Security surveillance is a basic work for everywhere. Most people actually don't pay much attention to privacy issues, but if face recognition is practically applied, the monitoring system and face recognition users can find the location of any person. It is indeed a good tool for combating crime and terrorism; but for ordinary people, it is a little bit terrible because they feel monitored.

Network Security Issue

At present, we basically take no measures to secure network security in the security field. Once a hacker gets a video recording or video stream, he can "borrow" someone else's monitoring system for face analysis as long as he has its own face recognition system.

Algorithm and principle of face recognition technology

Using Geometric Feature 


The use of geometric features for frontal face recognition is generally based on the extraction of important features of the human eye, mouth, nose and other important organ’s geometry as the classification basis. Roder has done a experimental research on the accuracy of geometric feature, but the results of the study are not optimistic.

Local Feature Analysis Method


Locality and topologicality are ideal characteristics for pattern analysis and segmentation. It seems that this is more in line with the mechanism of neural information processing, so it is important to work out a technique with this characteristic. This method has achieved good results in practical applications, and it forms the basis of FaceIt's face recognition software.

Principal Component Analysis


From a statistical point of view, that is to find basic elements of the face image distribution - the feature vectors of the face image sample’s covariance matrix, so as to represent an artificial face image. These feature vectors are called Eigenfaces.

Building Elastic Model


Elastic image matching is a recognition algorithm based on geometric features and gray distribution information for wavelet texture analysis. The algorithm works well because it makes good use of face structure and gray distribution information, and has automatic positioning and precise recognition function. 

Neural Network Method


Artificial neural network is a nonlinear dynamic system with good self-organization and self-adaptation ability. At present, the research of neural network methods in face recognition is in the ascendant. Lee et al. describe the characteristics of the face with six rules, and then locate the facial features according to the six rules, and input the geometric distance between the five features into the fuzzy neural network for identification. The application of neural network method to face recognition has an edge over the above, they are more adaptable and generally easier to implement.

Application Scenarios

Remote Authentication

To judge user's real identification through the online and offline mixed living body detection or comparing the public security identity image with the real person image, thereby completing the online user's identity verification.

Face Recognition Attendance

This application can record the face information quickly through face recognition, and users only need to let their faces scanned for entry to complete the authentication procedure. This technology can be applied in corporate, residential and other multi-scenarios, or for commercial uses, to improve the security, efficiency and user experience.

Security Monitoring

To conduct monitor in crowded public places such as banks, airports, shopping malls, markets, etc., to form automatic flow statistics, automatic identification and tracking of specific people.

Smart Album Classification

To automatically identify personas in the photo library through face detection function, then classify and manage them to enhance product user experience.

Facial Beauty

Based on face detection and key point recognition, it realizes interactive effects such as face beauty, special effects camera and patch function.

Face Attendance

To record the face pictures before the event starts, and you can sign in by scanning the face on the day of the event to improve the check-in efficiency.

Member Identification

Members do not need to present their membership credentials to the store, as long as they can complete the membership verification by scanning their faces, achieving "no card" identity confirmation and flow statistics.

Facial Recognition Gate

The face recognition function is integrated into the gate, it can solve the problem of forgetting to bring the work card and stealing others' card for entry.

Face Recognition Payment

To bind the face to the user's payment channel, and scan your face at the payment stage without showing the bank card, mobile phone, etc., so as to improve payment efficiency.

Face Login

To record a face image during the user registration phase to enable face login authentication in a scenario with high security requirements, therefore improving security.

0 comment

Leave a Reply

Your email address will not be published.