At the current historical turning point of interest rate marketization, booming Internet finance, and the formation of the three major factors of the new economic normal **** vibration, China's commercial bank business model is facing a brand new change. How to provide customers with better quality and safer service experience on the basis of refined operation and management has become the focus of competition among commercial banks. In recent years, cloud computing, the Internet of Things and artificial intelligence technology transformational development, related applications blossomed, the integration and utilization of "big data" resources and the intelligent development of the commercial banks to improve the "internal strength" of the cultivation method. Face information has the advantages of non-replicability, non-stealability, simplicity and intuition, etc. It is an important strategic resource that commercial banks should reserve and explore in the era of big data. With the popularization of technological changes and applications, the cost of building large-scale, distributed face database and recognition system is constantly reduced, and the accuracy of recognition is constantly improved. It is foreseeable that the potential value of face recognition technology in the field of commercial banking will be continuously explored and enhanced, and it has a wide range of application prospects in terms of guaranteeing service security, saving customers' time, enhancing customer experience, and integrating and mining data resources.

Overview of Face Recognition Technology

Face recognition technology is the process of matching faces with known identities in the database by means of extracting face information from a given static or dynamic image, with the goal of identity retrieval or verification. Due to the influence of light, expression, occlusion, orientation and other interfering factors, compared with other technical means based on ID cards, iris, palm prints, fingerprints, etc., the accuracy of face recognition technology is relatively low, but its acquisition method is the most friendly: without the cooperation of the person concerned, or even in the case of his/her unconsciousness, the acquisition and recognition of face information is completed. Therefore, face recognition technology has been a hot research topic in the field of artificial intelligence in the past forty years, and has gradually matured so far, and has been applied in the fields of anti-terrorism, security, access control, etc., and in recent years, it has begun to be promoted to the fields of education, finance, and so on.

According to different application scenarios, face recognition can be categorized into face recognition for 2D images, face recognition for surveillance video, multimodal face recognition for near-infrared, thermal infrared imaging or sketching, and 3D face recognition for depth information. For each of these data input types, researchers from academia and industry have proposed face recognition processing methods based on different assumptions, different models, and different disciplinary backgrounds. Summarized, these methods have similar processing steps, mainly including the following categories: First, face detection. The first is face detection, which solves the problem of "how many faces and where are they", i.e., detecting and determining the location of faces from pictures or videos, and separating them. The second is face tracking (for video faces). To solve the problem of recognizing faces "from where, to where", each detected face is tracked in each frame of the video, such as masking should be resumed after the end of the masking tracking, for example, two faces should be staggered without confusion. Third, face normalization. To solve the problem of "nose, eyes, mouth position on the right", the specific operation includes preprocessing, normalization, face calibration. Fourth, face recognition. That is, to solve the problem of "who is this person" (retrieval), "this person is not a customer" (verification).

In the establishment of the face database and recognition system, the need for face data training and modeling, if the database is dynamically updated will also involve online learning and so on; recognition of the face, the face to be identified and the database has a face to compare, to determine the degree of similarity between the two, and in accordance with the pre-set criteria for retrieval or verification. Face recognition has a variety of methods, such as: based on geometric features, based on subspace mapping dimensionality reduction, based on templates, based on models, based on neural networks and other methods.

Currently, methods based on "deep learning" have achieved high recognition accuracy in some algorithmic competitions, and are rapidly being applied in the industry. Deep learning does not refer to a particular algorithm, but is a general term for technical methods such as SparseCoding, RBM, and deep belief networks. As a class of neural network-based methods, according to cognitive psychology, its main idea is to simulate the signaling of human brain nerves. Unlike traditional neural network models with 2 to 3 training layers, deep learning can have up to 8 to 9 training layers. Therefore, when the idea was first proposed in 2006, the huge amount of training data and high computational complexity were beyond the capacity of the hardware at that time. However, due to the improvement of computer hardware performance, the advantages of deep learning algorithms in terms of accuracy quickly came to the fore. At present, Google, Microsoft, Baidu and other companies have set up specialized departments to research and develop deep learning technology, and a number of face recognition teams based on deep learning have emerged in the market. At present, deep learning-based methods have become an important development trend and direction in the field of face recognition technology.

In addition, some face analysis techniques have also been popularized and optimized with the development of face recognition technology, including the discrimination of expression, age, gender, and other attributes, which makes it possible to apply data mining clustering, classification, and other big data analysis applications based on these attribute information. Face recognition technology in practical applications, can also consider combining with other technologies or auxiliary means, such as combining depth information to achieve live body detection, to determine whether it is a real person or a photo.

Face recognition technology in commercial banks

Face recognition technology is currently mainly used in the public **** security field, such as: identification and tracking of terrorists, cloth control of high crime rate areas, airport security, driver license verification, video surveillance and so on. However, face recognition technology in commercial banks also exists a huge space for development. In the future, commercial banks can start from the security prevention and control and business promotion of two aspects of the face recognition technology in the bank landing comprehensive deployment and implementation.

Security prevention and control application scenarios

One of the difficulties of bank security is to complete real-time monitoring of multiple moving targets in dynamic scenarios. Face recognition technology in the bank and other densely populated areas can effectively achieve real-time multi-target online retrieval and comparison, the actual application of good results. Moreover, face information is easy to collect, difficult to copy and steal, and naturally intuitive, so face recognition technology can become the preferred choice of security prevention and control means for commercial banks. In the field of security prevention and control, the application scenarios of bank face recognition technology are as follows.

Business place personnel image control. In the business premises of commercial banks, face recognition can further ensure the security of bank operations through "camouflage identification". Through the identification of business premises in the facial mask (such as sunglasses, masks) of the personnel, the system can be real-time with the police database identity data for comparison, once the discovery of anomalies, you can quickly start the blacklist early warning mechanism or to take networking alarm measures. In addition, the collected facial photos of the suspects can be submitted to the public security authorities to provide powerful evidence for follow-up warning and case detection.

Identification of people in business banking areas. The extremely high requirements for security in the process of bank operations make its identity verification technology more stringent than other areas. For example, in the vault, the escort car, ATM machine money room and other special environments, many traditional authentication methods are difficult to meet the requirements, such as verification passwords are easy to be stolen, fingerprint identification can be copied, access cards are easy to lose. Face recognition technology with vivo detection can overcome these shortcomings and further enhance the security of bank security and confidentiality.

ATM machine intelligent recognition alarm. In the self-service equipment application scenarios represented by ATMs, face recognition technology also has a wide range of applications. For example: through the ATM machine built-in camera to identify the identity of the withdrawals, and the bank card owner information for comparison, to prevent the phenomenon of brush theft; identification of camouflage or intentionally cover the face of the identity of the person, and the police database for comparison, to ensure that the withdrawals of security. When the above situations occur, the system can trigger pre-set alarm rules to maximize the protection of bank customers' funds and personal safety. In addition, the face recognition system can also monitor the situation of customers' left behind belongings, real-time reminder, to enhance the user experience.

Business promotion application scenarios

The current application of face recognition technology in the field of banking business promotion is still in the ascendant, and the commercial banks are still in the active exploration stage. From the perspective of commercial bank business promotion, face recognition has the following applications.

Remote account opening and login. As an important process in bank account opening, face signing not only consumes customers' time, but also takes up the bank's human resources. By replacing the traditional visual recognition with face recognition, not only can we save time and cost, complete the whole process of operation from filling in personal data to opening a face-to-face account to picking up a card and activating it, and improve the user experience, but also can identify and relate the customer's identity and credit background within the whole network, avoiding the influence of psychological and experiential factors when manually signing the face. In addition, when customers remotely log in through mobile banking or online banking, face recognition can be used instead of the traditional password input operation to complete the functions of customers' bill inquiries, credit card repayments, personal card inter-card transfers, fixed and current inter-card transfers, and other personal fund transfers to avoid the phenomenon of stolen or forgotten passwords, and so on.

Personalized customer service. Currently, commercial banks are increasingly competitive, and the competition for customer resources has been transformed from product-oriented to service-oriented. Through customized personalized services to enhance customer experience will become an important means of commercial bank competition in the future. The use of face recognition technology can be very good to complete the identification of customers and accurate information search, when a customer enters the business network, face recognition technology can be used to quickly determine whether the customer is an existing customer of the bank, accurately obtaining the customer's name, age and other information, to facilitate the branch staff to draw closer to the customer's distance. In addition, through the extraction and analysis of the customer's past product purchases, transaction flow, business habits and other behavioral patterns, further targeted product promotion for customers, thus effectively enhancing the marketing success rate, and customers to achieve **** win.

Face recognition loan issuance. In the process of bank loan issuance, in order to effectively eliminate the phenomenon of fraudulent loans, malicious loan fraud, etc., can consider the introduction of face recognition technology for prevention and control. Based on the face information that loan customers have entered in the outlets, through the data *** enjoyment, you can realize the identification and verification of customers in the whole outlets, and really do the accurate correspondence between identity information and bank information, to realize the loan customer identity authentication informatization, intelligent, network management.

In summary, it can be seen that the application of face recognition technology in commercial banks has broad prospects. Combined with the actual business situation of China's commercial banks, it is recommended that banks prioritize the deployment and implementation of face recognition technology in the customer service field based on their own business development and business promotion. This is because: on the one hand, from the customer's point of view, face recognition technology is directly applied to customer service, can solve the most urgent needs of customers, give customers the most intuitive service experience, and help to quickly improve customer satisfaction; on the other hand, from the perspective of commercial banks, customer service-oriented face recognition application can quickly create profits for the bank, the effect is obvious, and can be for the face recognition in the bank's comprehensive landing to lay a good foundation. The company's business model is based on the idea that the company's business model is a good one.

Development Recommendations

Face recognition belongs to the applied research of computer science rather than basic theory, and the comparison and evaluation of different algorithms is based on experimentation and practice.

Currently, the industry's more common face recognition algorithm benchmarks are LFW (LabeledFacesintheWild) and FRVT (FaceRecognitionVendorTest). laboratory environment. In the database, there are multiple images of the same person and only one image. Since the images were taken in a normal day-to-day environment, they are highly practical, with a high coefficient of difficulty, and the results of the experiment are convincing. The FRVT was organized by the National Institute of Standards and Technology (NIST) to test the performance of face recognition algorithms in the industry. The benchmark uses a large database of 1.6 million people and is conducted from time to time, most recently in 2013, and is open only to industry.

Domestic commercial banks in the development of face recognition technology specific application landing, in addition to considering the LFW and FRVT two industry authority standards for reference, should also be based on the consideration of some of the actual situation, it is recommended that you can prioritize the choice of domestic technology products. First, because of the general convergence of international algorithms, the realization of foreign products is not significantly different from domestic compared to the effect, but the price is often higher than the domestic. Second, considering due diligence and personalized customization negotiations, domestic approach costs are much lower than foreign ones. Third, regulatory risk, given that commercial bank data belongs to core financial data, full consideration should be given to user data security under regulatory requirements, and legal risks caused by differences in domestic and foreign legal systems should be avoided as much as possible. Fourth, hardware constraints, if using foreign cloud service APIs, the problem of cross-border network delay may occur. Fifth, in terms of algorithm implementation, the foreign technology training data set of yellow pictures is less, which may affect the application effect. Sixth, considering the product after-sales service and technical support, the service response speed of domestic products has a greater advantage compared with foreign countries. In view of the above reasons, the domestic commercial banks in the application of face recognition technology can be prioritized from the domestic leading service providers to choose.

In addition, in some specific applications of the functional design, should also follow the relevant regulatory policies and industry norms. Take the remote account opening business as an example, the central bank issued in August 2015, "on the banking financial institutions to remotely open a RMB bank account guidance (draft for comments)" requires: "Banks use modern security technology means, the use of government department databases, the Bank's own database information, the commercialization of database information, through the customer information cross-verification, other Bank account cross-verification, telephone calls, mailing information, etc., to build a safe and reliable remote account opening customer identification mechanism; at the same time, according to the nature of the account opened to take the same legal person witnessed by different branches, through a third party to identify the identity of the customer, door-to-door identification documents, etc., to verify customer identity information, the depositary bank to assume the responsibility of customer identification. " The bank remote account opening business into the regulation, and clarify that the bank is the main body of responsibility for identifying customers. The central bank on December 25, 2015 issued the "Notice on Improving Personal Bank Account Services and Strengthening Account Management" will be the policy to the ground, which makes it clear that "to provide personal bank account opening services, conditional banks can explore biometrics and other safe and effective technological means as an auxiliary means of verifying the identity of applicants for account opening. " Face recognition technology, as an important member of the biometrics family, is expected to become a mainstream alternative. And with the development of business, more detailed specifications or guidelines may be issued again in the future. Therefore, commercial banks should pay attention to strengthen the learning of new policies and regulations, and maintain communication with relevant government agencies and regulators, so as to achieve compliance in the design and implementation of relevant processes and systems.

With the rapid upgrading of hardware equipment and the continuous development of algorithmic technology, face recognition technology has gradually moved from academic research to industry applications and has demonstrated strong vitality. Face information has the advantages of being easy to collect, difficult to copy and steal, and natural and intuitive. Face recognition technology provides new technical options for commercial banks in application modes such as security prevention and control and business promotion, and further develops the business operation mode. China's commercial banks can consider starting from business promotion services, based on the authoritative standards at home and abroad, prioritize the selection of domestic competitive service providers, step by step, all-round to promote the implementation of face recognition technology in commercial banks.