Face recognition anti-counterfeiting technology

Face recognition technology is a technology that extracts features of a face from a computer and performs authentication based on these features. Human faces are born with other biological characteristics of the human body. Their uniqueness provides the necessary preconditions for identity identification, and has the features of simple operation and intuitive results. Therefore, face recognition has a wide range of application prospects in areas such as information security, criminal detection, and entrance and exit control.

The principle of face recognition

　　 Face features

Because the face consists of the forehead, eyes, nose, mouth, chin, and other organs, the shape, size and structure of these organs vary, making the appearance of human faces vary widely. The geometric features of the human face are mainly reflected in the shapes and geometric relationships of key parts of the human face, such as the eyebrows, eyes, nose, mouth, etc. For example:

1. The thickness of the eyebrows and the vertical distance between the eyebrows and the center of the eye;

2. Eyebrow curvature;

3. The vertical position and width of the nose;

4. The vertical position and width of the mouth and the thickness of the upper and lower lips;

5. The shape of the chin;

6. The width of the face at the tip of the nose;

7. The width of the face between the tip of the nose and the eyes.

Face recognition identifies individuals by analyzing their unique shapes, patterns, and locations.

The main steps of face recognition

The basic framework of the automatic face recognition system is shown in Figure 1. First of all, image sensors such as cameras or scanners capture face images and have fewer constraints and simpler operations than iris, finger, and hand-vein vein acquisition. Then, the image is pre-processed to remove or reduce the interference of the image to be processed, such as illumination, imaging system, and external environment as much as possible, to provide high-quality images for subsequent processing. Then the face detection algorithm is used to detect whether there is a valid face in the image, and if so, the status information such as the position and size of the face is given, and the face image is extracted and extracted (face detection is a very critical step and is mainly subject to light, Noise, posture, and occlusion and other factors). Then, to extract the features in the face image, how to extract stable and effective features is the key to correct identification. Finally, the classifier is used to classify and identify the extracted features. The accuracy and time of recognition are crucial indicators. The core of this process is to select appropriate face representation methods and matching strategies.

Figure 1 The main steps of face recognition

Development and Application of Face Recognition Technology

From the late 1960s to the early 1970s, face recognition research has just begun. The earliest researcher was Bledsoe, who established a semi-automatic face recognition system, characterized by parameters such as pitch, ratio, etc. of face features. The early face recognition method has two major features:

1. Most recognition methods are based on components. They use the geometric features of human faces for recognition. The extracted information is the main facial feature information and the geometric relationship between them. This kind of method is relatively simple, but it is easy to lose the useful information of the face, so that the ability to recognize in the case of changes in perspective, expression, etc. is poor. In view of this situation, there has emerged a template matching method with better performance. That is, according to the degree of similarity between the face template in the image library and the face template to be recognized in the gray level, face recognition is implemented. Such a method is in a certain period of time. Within the mainstream.

2. Face recognition research is mainly face recognition under strong constraints. Assuming that the image background is single or no background, the face position is known or easily obtained, and the image processing effect on the real scene is not good.

Since the 1990s, with the rapid improvement of the performance of computer hardware and software, as well as the high requirements for face recognition capabilities, the development of more robust face recognition methods has become inevitable. So a whole-based identification method came into being and quickly became the focus of research. This method makes full use of the topological relationships between the individual feature points of the human face and the information of each organ itself, which can avoid the operation of extracting the facial local features and improve the robustness of the recognition. As a result, in the area of ​​face recognition research, there has been an overall approach based on components and component-based approaches.

Since the mid-1990s, face recognition methods have evolved toward the combination of overall recognition and component analysis. Researchers began to gradually realize that face recognition algorithms must be able to make full use of various features of human faces, and integrate features such as topological features of the human face, local gray features, and global gray distribution features. Therefore, many new algorithms have emerged. These algorithms combine the original single algorithms to complete the recognition of faces. One of these is the deformable model method of grayscale and shape separation.

In the late 1990s, some commercial face recognition systems began to enter the market gradually, and face recognition technology became one of the most important means of international security prevention. However, these technologies and systems have a certain distance from practical application, and their performance and accuracy need to be improved.

In recent years, although the performance of face recognition method has been improved, it still has a certain gap with people's requirements. The existing methods are sensitive to changes in conditions such as light, age, expression, posture, and distance. When certain conditions change, the recognition effect is not ideal. At present, the face recognition technology can only be used in some occasions where the recognition accuracy rate is not high.

At the present stage, face recognition technology from the initial recognition of the background of a single positive grayscale image, through the study of the face recognition of multiple gestures (front, side, etc.), has been developed to be able to dynamically achieve face recognition, is currently to the three-dimensional The direction of face recognition develops. In this process, the image involved in face recognition technology is gradually complex, and the recognition effect is continuously improved.

Although face recognition research has accumulated a wealth of experience, the current recognition technology still cannot effectively handle and automatically track human faces such as in complex backgrounds. At the same time, unlike other disciplines, the face recognition technology incorporates the theories and methods of digital image processing, computer graphics, pattern recognition, computer vision, artificial neural networks, and biometric technologies, and requires researchers to be perfect. Knowledge system and rich experience. In addition, the complexity of the human face and its environment, such as the change of conditions of the expression, the posture, the ambient light intensity of the image, and the covering on the human face (glasses, whiskers, etc.), will make the face recognition method robust. Sex is greatly affected. Therefore, face recognition technology remains a challenging subject in the 21st century.

Face recognition technology has now achieved initial application in many fields.

In the field of public safety, based on the face recognition access control system, the face information stored in the database is compared with the face information of the visitors to identify, match, and enable the access control system; the network video based on face recognition The monitoring system, which combines the collection of face data and digital monitoring, more effectively plays the role of post-supervision, has been widely used in network security monitoring; face recognition security system, for people entering an environment The identity is verified and confirmed; the driver's license verification is based on face recognition and so on. Following the September 11th incident in the United States, face recognition has become one of the most important means of international anti-terrorism and security prevention.

In the civil and economic fields, face recognition technology has important application value in the identity verification of bank card holders and identity verification of social insurance. In addition, face recognition technology has also been applied to the management of computers and some electronic products (such as mobile phones), and has received very good application results.

In areas such as home entertainment, face recognition also has some interesting and useful applications, such as smart toys that can identify the owner, housekeeping robots, virtual game players with real faces, and so on.

Through half a century of continuous evolution and development, face recognition technology has had a number of mature theories and effective algorithms, and domestic research institutions have also introduced some effective commercial systems, providing a large number of databases for many researchers. Sharing, it should be said that face recognition technology has completed pioneering work and reached a certain degree of maturity. However, even the most mature face recognition system in the world can only achieve its intended function according to the condition of the general expression of face recognition, that is, to give a still or video image of a certain scene and use its own face. The database determines the identity of one or more people in the scene, that is, it can only basically reach the applicable level under the condition that the collection conditions are ideal and the users are more cooperative. Under non-ideal conditions (such as expression changes, age changes or makeup, wearing glasses hats, etc.), although humans can easily recognize faces in complex backgrounds, they are quite large for computer automatic face recognition systems. The challenge, its recognition rate is far from reaching the ideal applicable level.

On the other hand, because the face is an external feature, the desired face image can be obtained easily without authorization, which provides the possibility of counterfeiting. With the development of science and technology, high-tech attack methods such as simulation caps, holographic projections, and face tracking continue to emerge. In March 2016, the Stanford University research team released a face tracking software Face2Face developed by them. It can use ordinary cameras to capture the user's actions and facial expressions, and then use the software to drive the target people in the video to make the user and the user. The same actions and expressions, such as the use of Face2Face users can control Putin, Obama, Bush and other great people in the video to make any user wants strange expressions, the effect is extremely realistic. In the future, with the popularity of high technology, the cost of face recognition attacks will continue to decrease, which will be a huge challenge for face recognition technology.

- Excerpted from Chapter 6 of "Anti-Counterfeiting Technology in Life"