Opportunities and challenges of biometrics in the AI era
Author: huifan Time: 2019-12-06
From key unlocking to fingerprint unlocking to face unlocking, the development of biometric technology has begun to focus on liberating human hands. At the same time, biometrics is beginning to serve every aspect of people's lives. The era of face brushing is the most straightforward example.
Baidu explained that biometrics is a technology that realizes identity authentication based on human physiological characteristics (face, fingerprint, iris, etc.) and behavior characteristics (posture, motion, emotion, etc.). When performing human identity authentication, it mainly uses computers and optical, acoustic, biosensor, and biometric principles to closely integrate high-tech methods, and uses the inherent physiological and behavioral characteristics of the human body to perform personal identification. At present, the commonly used biometric technologies mainly include: face recognition, fingerprint recognition, iris recognition, behavior recognition, and gait recognition.
Face recognition: Face recognition, also known as facial recognition, is a biometric technology for identifying people based on their facial features. Because face recognition involves many organs and large area, the recognition technology is not only complicated and susceptible to interference from many factors, such as human expressions, gestures, etc., and the micro-expression recognition and emotion recognition mentioned now Technology belongs to the branch of face recognition technology.
Relevant reports show that with the increase in government and user security initiatives, the increasing use of mobile devices, and the growing global demand for sound fraud detection and prevention systems, the size of the face recognition market is expected in 2019. It will be USD 3.2 billion, and the market size will reach USD 7.9 billion five years later, 2024, with a compound annual growth rate of 16.6%. In other words, face recognition technology has become the most mainstream biometrics technology.
|Biometrics||Convenience||Intuitive||accuracy||effectiveness||Security Level||Long-term stability||Identify equipment costs||universality||Counterfeit||Possible interference|
|Fingerprint recognition||Higher||Medium||High||Medium||Medium||Higher||Medium||Medium||Medium||Dirty, greasy, skin abrasion|
|Face recognition||Extremely high||Higher||High||High||High||Higher||Medium||Higher||low||Light,Occlusion|
|Iris recognition||Medium||Medium||Extremely high||Medium||Extremely high||Higher||Higher||Higher||low||Contact lens|
|Speech Recognition||High||Medium||Medium||High||Higher||Medium||Lower||low||Medium||Voice, cold|
From a technical perspective, the face recognition system is mainly composed of four parts: face image acquisition and detection, face image preprocessing, face image feature extraction, and matching and recognition. At present, with the improvement of various computing power and algorithms, and the unstructured data is gradually converted to a structured level, the accuracy rate of face recognition algorithms has reached an average of 99.69%, that is, the face recognition technology has reached a high level. And in China. The "Four Little Dragons" of Chinese face recognition, represented by Yuncong Technology, Shangtang Technology, Yitu Technology, and Deaf Technology, have already appeared.
From the application level, in China alone, with the promotion of Skynet project, Xueliang project and national policies, China has become one of the countries that have benefited most from face recognition technology, and software and hardware products based on face recognition technology have also continued. In the update, such as face ramps, face recognition mobile phone unlocking, face recognition access control and attendance, integrated authentication and verification, face recognition arresting fugitives, pedestrians running through the red light grabbing system, Alipay face recognition payment, hospital online card application, etc. And so on, various "smart +" industries, such as smart security, smart education, smart hospitals, smart finance, smart transportation, etc., have made face recognition technology the most mainstream technology.
Although the world has accelerated into the "brush face era", the development of face recognition still faces many challenges.
● 2D face recognition technology
Due to the limitation of deep data loss in 2D information, the real face cannot be fully expressed, so its robustness to complex faces is not good, and there are also many difficulties in practical applications, such as cross-view (cross (Pose, cross-age, before and after makeup, etc.) face recognition, low-resolution faces, deep learning (sample deception), face anti-deception, complex scenes (strong and weak lighting, blur, occlusion, etc.), external changes, computing power Limitations, fine-grained classification (non-human detection and agent detection), and more.
● 3D face recognition technology
At present, 3D face acquisition technology includes binocular technology, structured light technology and TOF technology, and 3D live detection can effectively distinguish attacks such as fake pictures, videos, masks, etc., but there are also many challenges in algorithm and hardware. In terms of algorithms, there are two major issues: how to obtain high-quality 3D face images (depth images, point cloud images, grid images, and how to solve the impact of factors such as face poses, expressions, and occlusions); at the hardware level, The challenges include lighting, distance, accuracy, complexity, and more.
In addition to the above problems, face recognition technology also faces problems such as multi-modal technology fusion issues, artificial intelligence chip monopoly, and data privacy issues.
Although face recognition is the most popular biometric recognition technology in the market, fingerprint recognition is still the most widely used recognition method. Fingerprint recognition is mainly based on the comparison of the detailed feature points of different fingerprints for identification. The process can be summarized as fingerprint collection, fingerprint feature extraction, and fingerprint matching. At present, the development of the fingerprint recognition industry can be summarized as follows:
● The fingerprint identification industry includes chip design, chip manufacturing packaging, module manufacturing, and complete machine manufacturers, etc., and the industry chain is becoming more complete.
● In recent years, more and more smart phones have begun to adopt fingerprint recognition technology. In 2013, iPhone 5S and Huawei smartphones with fingerprint recognition were launched;
● With the promotion and popularization of full-screen mobile phone solutions, fingerprint recognition sensors have begun to be rear-mounted or built under the display.
● Fingerprint recognition still occupies the highest share among various biometric technologies. In 2017, the global fingerprint recognition technology scale was approximately 7.456 billion US dollars, a year-on-year increase of 7.71%.
It is foreseeable that with the development and improvement of related technologies, fingerprint recognition will be widely used in smart terminals, ID cards, motor vehicles, home furnishings and more. However, there are still many pain points that need to be resolved in fingerprint recognition
● Quality evaluation: Fingerprint recognition depends to a large extent on the quality of the fingerprint image obtained. Solving the problem of fingerprint quality evaluation is very important
● Data compression: The amount of data stored in the fingerprint database is large, and all fingerprints need to be matched and scored during fingerprint recognition. How to compress the data to achieve fast matching is worth considering.
● Authentication: The fingerprint information is easy to be counterfeited, and the fingerprint is low and difficult to distinguish.
● Privacy and security: At present, most fingerprint recognition systems do not perform irreversible encryption. Once the fingerprint information is leaked, it will cause loss to users and even threaten public safety.
● Fingerprint acquisition hardware and chip: the current under-screen fingerprint recognition technology is mainly used in OLED panels, which is difficult to do in liquid crystal panels (LCD)）
In addition, fingerprint identification does not actually have a lifetime "uniqueness" and "everyone's fingerprint is not unique": people with blood relationship will have very similar fingerprints, and their fingerprints change with age and skin conditions , Fingerprint shape will also be affected a lot.
The iris recognition technology is based on the iris in the eyes for identification, and is applied to security equipment (such as access control, etc.), and places with high security requirements.
Iris recognition has high stability, accuracy, and security. Compared with fingerprint recognition and face recognition, it has obvious advantages. However, because iris recognition technology is more difficult, it has higher production costs compared to other biometric technologies, and it has The requirements for recognition distance are high, and these factors have prevented them from entering the general consumer market to some extent.
In addition, the current development of China's iris recognition technology is still limited by many factors, one is the acquisition technology and equipment, and the other is the national policy factor. At present, China's iris collection behavior is still in its infancy, and to achieve the same wide application as face recognition, it needs the strong support of national policies;
In addition, because iris recognition requires a more special acquisition lens than face recognition, and it is susceptible to external environmental interference, its penetration rate needs to be improved.
However, fortunately, with the support of relevant departments, China's iris image recognition and identification work has begun, and the iris recognition market will definitely be promising in the future!
Behavior recognition: With the continuous maturity and development of face recognition, fingerprint recognition, and iris recognition technology, biometric recognition technology has gradually evolved into behavior recognition technology that currently focuses on gesture estimation and motion recognition.
In a nutshell, behavior recognition is an algorithm that structures the main activity skeleton of a person, defines a variety of behaviors based on the person's motion trajectory, and forms an action system through a deep learning algorithm that can be efficiently recognized by the system . This recognition technology captures the skeleton of a person by shooting with a camera, and analyzes the movement trajectory of various movements of the person to determine what kind of movement the movement trajectory belongs to. Then the calculation is performed through the background server. Once the management actions set by the system are matched, the system will immediately give an early warning to achieve the purpose of early warning.
At present, thanks to several advantages of behavior recognition technology: many actions can achieve zero false positives, can accurately identify human abnormal behavior in the scene, high server recognition and analysis efficiency, and the same camera can analyze N Abnormal behavior, this technology can analyze emergency behaviors such as emergency call for help, fighting, high-altitude throwing, crowd watching, etc., and can give early warning.
However, the technology is always flawed. Although behavior recognition technology gradually brings security closer to active defense, the current development of behavior recognition technology also faces algorithmic challenges and hardware challenges.
● Algorithm challenge: Lack of end-to-end models, diversity of human poses and movements, complex scenes, lack of well-labeled large data sets, individual differences (different people's performance in unified actions)
● Hardware challenges: There are certain requirements for high-precision, miniaturized sensors and high-computing, low-power chips.
Like behavior recognition, gait recognition is also a rising star in the field of biometrics. Gait recognition aims to identify people by their walking gestures. From an anatomical perspective, the physical basis of gait uniqueness is the difference in the physiological structure of each person, different leg bone lengths, different muscle strengths, different centers of gravity, and different motor nerve sensitivities. Determines the uniqueness of the gait. In addition, due to the technology's support for long-distance identification, no need for hard cooperation, and strong environmental adaptability, it has begun to enter the security, transportation, industrial and other industries for relevant applications.
From a technical perspective, compared to static biological features such as fingerprints, face, palm prints, veins, gait is a dynamic feature, so it is more complicated in the recognition process. The entire process of gait recognition is divided into four major steps: collection, analysis, extraction, and comparison, but in fact each link faces challenges. For example, the collection of data samples, how to obtain data, and how to build a database of gait recognition? How to segment the foreground and background after obtaining the data to make the recognition more accurate? At the stage of feature expression, how to solve the problem of cross-view recognition and so on.
As a new type of biometric device technology, gait recognition needs to move from the laboratory to a commercial scene. Gait recognition must not only tackle one by one in terms of recognition accuracy, recognition speed, technology application cost, convenience, etc., but also It is highly integrated with the industry and has gained industry recognition. ......
Relevant reports show that by 2020, the global biometrics market size will exceed US $ 25 billion, and China's biometrics market size will reach 30 billion yuan.
Not only the five major recognition technologies mentioned above, biometrics also include voiceprint recognition, eyeprint recognition, retinal recognition, vein recognition, human body recognition, and more. The challenges faced by all these biometric technologies come from the three aspects of algorithms, hardware, and laws and regulations: at the algorithm layer, large-scale data sets that need to be well labeled to meet the interpretability of deep learning models and the complexity of practical application scenarios; At the hardware level, we need to focus on the design and manufacture of sensors, the design and manufacture of chips, and the research and development of real-time computing of mobile devices. Finally, the state and relevant governments must start with laws and regulations and formulate relevant policies to ensure user privacy protection and unification. Industry Standard.
Summary: It is undeniable that from fingerprint authentication to face recognition and iris recognition, biometric technology is entering the "visual era". It is true that technology is constantly improving, but behind this "visual age", multiple biometrics is the king of future biometric technology development.
Compared with single biometrics technology, multiple biometrics technology that combines face recognition, behavior recognition, gait recognition with passwords, fingerprints, iris and other methods will bring higher reliability and accuracy, while maximizing protection User privacy issues.
Looking forward to the future, multiple biometrics technology will move toward a wider and wider application field and market!
Facebook: Huifan Technology