Face recognition in finance has two main uses: One is identity verification, which is one-on-one. In fact, you already know the identity of the person to be verified that you currently operate, and then only need to conduct an information check between this identity and this person. Another kind of what we call it is called 1 to N. It is used in large-scale scenario applications. For example, if you patronize a service, you may have a VIP card. Then we don't need to name a name, and we don't need to use a VIP card. , To brush his face at the door, he knew which one he was, and automatically deducted money after the transaction was completed. Such a scenario, because no one actually speaks who is who, but in the case of 1:N, it can search for unclear target objects and find the correct target object.
In fact, in the business scenario of remote identity verification, the technology used is not only face recognition. When any technology lands on a practical application, it is actually an application of a comprehensive technological solution. For example, the most commonly used assistive technology in the identification of one-on-one: identity text content identification (IDR OCR). There are still some industry-specific issues to solve. To solve the problem with mobile phones, one problem that needs to be solved is to prevent users from using photo forgery to brush their faces. We call it live recognition. We need to determine whether it is a real person or a person to brush their faces. The photo is still a video.
Many people may ask, what level of face recognition technology has reached? In fact, such certification payments in the financial industry now have very high security. From the point of view of face recognition alone, it is now relatively common to control the pass rate and convenience at 90% or 95% to see misrecognition and security.
Some people say why it is not set to 99%, we can set it to 99%, but 99% will have an impact on security. These two indicators will conflict, and the financial industry requires that the higher the security is, the more acceptable the convenience is. it is good.
Voiceprint fingerprint iris and other multi-biometric fusion authenticationDoes this mean that face recognition can already completely replace bank card passwords and even achieve a higher level of security? Not too.
According to Li Fan, head of voiceprint technology in the Feiyun Platform Division of the University of Science and Technology of China, as far as face recognition technology is concerned, under the latest DFDR algorithm, the face resolution capability has reached 99.47%. However, there are still some challenges in practical applications - such as when faced with a South Korean lady. Of course, voiceprint technology also has some application challenges. For example, voiceprints are easily deformed and are susceptible to physical conditions, age, emotions, etc. In practice, they may also be affected by recording equipment, so as a single organism Each feature has its own instability.
In order to cope with a high security requirement in the payment field and also to address the instability of a single biometric feature, the industry has proposed a multi-biometric fusion authentication solution that combines the two biological features from different dimensions.
Earlier, the international biometric organization conducted a systematic statistical analysis of the biometrics currently in use on the market from four levels of user level, independence, cost, and ease of use. The analysis included palmprint recognition, signature recognition, and Fingerprint recognition, speech recognition, iris recognition, retina recognition, face recognition, and warm-spectrum recognition.
The ideal biometric system of course is to have high uniqueness, high ease of use, low degree of interference to the user, and low cost. From here, we can see that in the scope of human cognition voiceprint recognition, face recognition Fingerprint recognition, iris recognition, and palmprint recognition in the context of technology and human cognition can play an important role in their respective fields.
Among them, voiceprint recognition technology and face recognition technology can be used as a convenient form of identity authentication. It is inseparable from the characteristics it possesses. First of all, these biometric features are inherent, and it is unique and not easily copy. Second, these features are easy to collect, and the acquisition is more concealed. It can also be operated remotely.
In the case of voiceprint technology, voiceprints automatically recognize the speaker's identity based on the speech parameters that reflect the speaker's psychological and behavioral characteristics in the speech waveform. Each person has a unique voiceprint, no matter how similar they are. This is formed by each person's vocal organ during its growth. This uniqueness can uniquely determine the identity of a user.
At present, the voiceprint recognition mode supports voiceprint freedom, dynamic digital password, open text password, and fixed text password. In financial payment, the dynamic digital password is currently the most widely used voiceprint recognition mode because of its ease of use. Of course, the other three modes also have corresponding applicable scenarios.
The main uses of voiceprint recognition in this field are divided into two types, similar to face recognition. One is the confirmation of voiceprints, the one to one comparison, and the second is the comparison of voiceprint identification to one-to-N.
It is believed that the most curious thing for everyone is the process of establishing and pairing this voiceprint model. What exactly is the principle and the ability to quantify a person's voice using algorithms and code. Li Fan told Lei Fengwang (searching for the "Lei Feng Net" public number concerned) that, as far as HKUST flight information, it uses a hybrid Gaussian-universal background model (GMM-UBM) and is a speaker system based on a high-speed hybrid model. The main algorithm in people recognition.
However, biometrics based on physiological characteristics have the following congenital defects: First, the cost is high, usually requires special hardware support; Second, easy to leak and counterfeit, low security; Third, improper application of serious consequences, once the identity of the leaked Information can hardly be changed. Therefore, in addition to biometrics based on the human body's own biological objects for security monitoring, the industry is also studying methods based on key biometric features such as keystrokes.
The keystroke process may seem simple, but in fact it is very complicated and requires a lot of muscles with a high degree of coordination. Over time, a fixed habit is formed, which is reflected in the time and pressure of keystrokes.
Researchers said that in specific applications, keystroke feature recognition of free text and keystroke feature recognition of fixed text are also different. For fixed texts, the length is generally short, and the keystroke characteristics expressed by the user when typing fixed texts are greatly affected by the restriction conditions, and the final keystroke morphology is also closely related to the restriction conditions. In this case, the keystroke format can be directly stored and compared... For free text, its content and length are not limited, and the influence of the restriction condition on the final keystroke form may be very limited. In this case, the keystroke feature samples of the user can be identified through two aspects: one is the restriction condition; the second is the traditional free text keystroke feature recognition algorithm. Constraint conditions are compared with database storage; free text key signature recognition algorithms also have many ready-made solutions, such as Manhattan distance, Euclidean distance, neural network, relative entropy, and so on.
However, some users think that the keystroke feature is a false proposition. Because different keyboards implement different standards, the keystroke and elasticity will be affected. In fact, every technology has its own flaws. What we can do is constantly optimize it or make up for it in many ways.
Related Reading:
When finance and technology are extremely hot, what can be done to control the wind under payment scenarios? (on)