Source of this article: Wulian Media
Author: Vior.Liu
Internet of Things Exchange: Sign up for the Ten Thousands Event
From the beginning of this year to the present, ToF sensors have always been the technology that Apple, Samsung, GD, AMS and other sensor companies and smart hardware companies pay attention to. The most widely used field of ToF sensors is face recognition.
Almost everyone has used face recognition in application scenarios such as government affairs, access control, payment, etc. According to the survey, 90% of people have used relevant technology. Although the convenience is high, the security problem of face recognition is born from technology. Up to now, there have been doubts, not just face recognition, but biometric technologies such as fingerprint recognition, voice recognition, iris recognition, and vein recognition have always been hot topics in the AI field.
However, the security of face recognition is frequent thunderstorms, causing the security of biometrics to attract attention. So what kind of business is biometrics?
Fingerprint recognition and face recognition are not the best biometric methods
From the perspective of application trends, fingerprint recognition and face recognition are currently the most widely used biometrics, but from a technical point of view and security, neither of them can be regarded as ideal biometrics.
In the field of biometric recognition, it is divided into first and second generation recognition technologies. The fingerprint recognition, face recognition, iris recognition, palmprint recognition, DNA recognition, signature recognition, voiceprint recognition, and gait recognition mentioned above all belong to The first generation of biometric technology. Vein recognition (divided into finger vein recognition and palm vein recognition) and retina recognition belong to the second generation of biometric technology.
From the perspective of security and technological iteration, the second-generation biometric technology has more advantages than the first-generation recognition technology.
The difference between the two generations of technologies can be distinguished from the visibility of features and in vivo recognition.
The visibility of features , if the physiological or behavioral features can be seen by the naked eye, making it easy to be imitated and copied, this kind of biometric technology based on physiological/behavioral features can be called the first-generation biometric technology, fingerprint recognition, face Recognition, iris recognition, palmprint recognition, signature recognition, voiceprint recognition, and gait recognition. The behavior or characteristics of these recognition technologies are visible. Fingerprints, faces, and palmprints can be achieved through some techniques or hardware and software. Tools are extracted or copied, while voiceprints, signatures, and gait can be deliberately imitated to obtain highly similar behavior characteristics.
Living body recognition. "Living body recognition" refers to live body detection and recognition that cannot be replaced or simulated by external forces. That is to say, if you copy or 3D print a face model, the prosthesis recognition can only be regarded as the first-generation biometric technology. The living body recognition can judge whether it is a real living body, and the living body that meets the requirements of visibility and living body recognition can be regarded as the second-generation biometric technology.
For example, in vein recognition, the veins are hidden in the fingers or palms and are not visible. In addition, the vein recognition technology uses infrared light of a specific wavelength to irradiate the human body. The hemoglobin in the blood in the skin and blood vessels under the skin has different reflections of infrared rays. Characteristics, real-time acquisition of blood vessel images, feature comparison and matching with stored images to achieve identity authentication and identification, so fake fingers and finger images cannot be identified through veins. Vein recognition meets the two requirements of the judgment and belongs to the second-generation biometric technology.
Retina recognition is similar to vein recognition and belongs to the second-generation biometric technology. The retina is the blood cell layer at the bottom of the eye, and its characteristics are not obvious; secondly, if there is no blood flow or it is a non-living body, the retina recognition cannot pass. However, retinal recognition requires laser to irradiate the back of the eyeball, which may cause damage to the eyeball, and it is also difficult to reduce the cost. This article does not make comparisons.
As can be seen from the figure below, fingerprint recognition and face recognition are not the best biometric methods in terms of multiple technical indicators.
However, from the perspective of cost, convenience, and acceptance, fingerprint recognition and face recognition are indeed the best ways to popularize biometrics, and now they are worthy of the name in terms of market share. According to the latest data from Transparency Market, fingerprint recognition accounts for 58% of all biometrics, and face recognition accounts for 18%, ranking the top two.
However, due to the stimulus of the epidemic this year and the growth of non-contact economy and technology, the vein recognition market with higher security is also constantly entering people's attention. In July of this year, Apple’s application for “Vein Matching for Difficult Biometric Authentication Cases” mentioned that the use of facial veins for face recognition, this technology has more advantages than finger veins and palm veins. For complex vein structure, it is more difficult to imitate. Similar to Apple, in January this year, Amazon also conducted a hand payment test by developing palm vein technology.
In terms of market growth, the market share of vein recognition in 2020 has increased by approximately 4 times compared with 2015, and the market size is approximately US$2 billion. The author believes that due to the gradual installation of near-infrared sensors in smart phones, the market growth rate of vein recognition will further accelerate.
In Japan, finger vein technology has been widely used, such as bank ATM machines, PCs with access to a large amount of personal information, access control and attendance management systems, safe management, copier management, electronic payments and other fields that require identity authentication. At the same time, in the above fields, almost all patents are held by Japanese companies, including Hitachi and Fujitsu.
Multi-modal, multi-sensor fusion is the future of biometrics
Dialectically, although the safety of vein recognition is high, its popularization also presents huge challenges.
First of all, blood vessels are 3D, and imaging patterns at different positions and angles are quite different. Especially in finger vein recognition, a slight deviation will cause misunderstanding or misunderstanding, which greatly affects user experience. Secondly, cost is also the most important factor restricting the large-scale promotion of vein recognition products.
Combining fingerprint recognition and face recognition, there is currently no single biometric technology that can perfectly adapt to market needs. At the same time, due to the complexity of the actual identification system construction and application environment, single biometric identification will have different problems. For example, the near-infrared sensor used for vein recognition and the ToF sensor used for face recognition will have noise in the process of collecting data, which will affect the accuracy of the data; applicable people are not universal, such as people with disabilities; and the aforementioned fingerprints and The security problem of face recognition is easy to be copied. Therefore, single biometrics will have limitations in actual application scenarios.
Based on this, recently in the field of biometrics, multi-modal and multi-type biometrics fusion technology is regarded as the future trend. Smart phones, smart door locks and security fields have already appeared in the application of multi-biometric technology integration.
This trend is also compatible with the current multi-sensor fusion technology advocated by sensor companies. This point needs to be pointed out that whether it is the proposed multi-sensor fusion or multi-modal biometric technology, essentially it does not use different sensors or biometric algorithms to obtain corresponding results, but integrates multiple data through fusion algorithms. Determine the final recognition result or data. This processing method is more conducive to the processing speed and security of biometrics and IoT data.
It can be seen that the multi-modal biometric fusion first needs to be built on the basis of sensor fusion for algorithm fusion. Looking at the country, although it is widely used in fingerprints and faces, it is relatively backward in the second-generation biometric technology such as vein recognition, which is more secure and more intensive. Therefore, the development of domestic multi-modal biometrics still takes time .