Research on brush face payment system based on internet artificial intelligence

Abstract

The facial information of a person has constant invariance, and at the same time, it is an ideal basis for identifying identity information through differences between different individuals. The use of face recognition in third-party payment platforms will greatly improve the efficiency and security of payment. Based on this, the paper studies and designs the third-party payment system for face recognition in the context of Internet artificial intelligence to improve the convenience and security of face payment.

Keywords

Internet artificial intelligence face payment face recognition third party payment system

1 Introduction

With the advent of the electronic information age, the number of transactions completed through the Internet has increased day by day, and cash payments have gradually faded out of people’s horizons. The popularity of online banking, credit cards, and Alipay has changed the capital flow pattern of traditional transactions. The world’s largest online payment company is PayPal in the UK. At present, PayPal’s “Face Recognition” payment function requires only one smartphone to make payment. Freed the bondage of credit cards and created an era of cordless transactions [1].

The transaction payment page displayed by the mobile phone Taobao during the transaction is no longer the Alipay password input interface, but the face scanning interface. The user only needs to point the mobile phone to the face for scanning when making the payment. This payment method replaces the traditional password input payment. Alipay official said that the face recognition payment function in the international public test accuracy rate of up to 99.596, using the world’s latest “interactive command+continuity judgment++3D judgment” technology. When performing face recognition payment, the system does not require the user to upload related photos in advance. When scanning the face, the system will apply to the National Citizen ID Card Number Search Center to perform the identified facial features and the photo library collected by the Public Security Department’s inquiry center. Owing to losing the detailed information, the low-resolution problem in face recognition degrades the recognition performance dramatically. To overcome this problem, a novel face-recognition system has been proposed, consisting of the extracted feature vectors from the multiple-size discrete cosine transforms (mDCTs) and the recognition mechanism with selective Gaussian mixture models [2 , 5].

Since the introduction of the face payment function, it has been widely welcomed by the public as a new payment system based on face recognition technology. The face payment system has been used by many companies such as Tencent, Baidu and Alibaba. However, due to the certain instability and reproducibility of face recognition technology in China, when face recognition is used for payment function, there is a lack of certain supervision, the behavior lacks standard standards, the user’s personal account funds and personal information are used. There are more risks in the face payment system. Based on the research background, this thesis takes Alipay as the object, and studies the face recognition process and system design implementation of the face-based payment system based on the Internet background. At the same time, the face recognition algorithm is improved to achieve the ideal payment security effect.

2 Alipay artificial intelligence brush face payment system technical framework

In the design process of Alipay framework, the system should divide the whole design platform into three levels: operation and maintenance platform, which mainly provides the scalability of basic resources, including IDC, virtualization, database, storage, and network. The stability of the underlying system platform is effectively guaranteed. In addition, the high-availability distributed transaction processing and scalable distributed transaction processing methods and high-quality service computing capabilities ensure the technology platform in resource allocation and remote access. The initiative in control provides a basic middleware operating environment to shield the complexity of the underlying resources; the business platform provides high-availability payment services anytime, anywhere, and provides a secure and easy-to-use open payment application development platform. Its structure is as follows [4]:

2.1 Logical data center architecture

At present, the challenge that Alipay needs to face is: the storage capacity of the system is increasing day by day, so the computer room, database, network, and server are gradually growing. This also causes some serious problems, such as the scale of the system. Increasingly, the system has gradually become more complicated. The original design based on the scalability of the point cannot achieve the current purpose, so we should prepare a more comprehensive and scalable plan, which can be extended according to a single level. According to the needs mentioned above, the logical data center architecture emerges as the times require. The main idea is to transform the data horizontally into the terminal and access layer. The system is divided into dozens of units from the access layer. The units have their own characteristics: all units are not open to the outside world, including the exchange of various types of storage between systems; the real-time data of each unit is independent and not shared. The member or configuration class can share the data with low latency requirements; the communication between the units is unified and controlled, and the asynchronous message is taken as much as possible.

2.2 Full routing layer

The full routing layer provides developers with the API functions necessary to call the underlying service and drive the interface. Through the application architecture, developers can easily call the functions and services that have been implemented, and make the development work simple and efficient by reusing the code. Correspondingly, through the same framework and rules, all applications provide interfaces to other applications and services when calling other applications and interfaces of services. P

2.3 Alipay brush face payment data interaction

When the order information is created and signed at the same time, the merchant client will follow the interface mode of the development package paid by the wallet, and the signature result formed by the program and the data unified to be sent to the development package; the data transmission will already be The created data continuum is delivered to the development package; the development package then processes the request, which processes the data according to the business regulations and then delivers it to the Alipay server, which after obtaining the collection The test will be started. After a series of tests are completed, the data will be processed. In the case of transactions that have already been processed, Alipay will return the data to the merchant’s server and its application in two ways. The Alipay server first gives a notification to the page path created by the merchant at the same time as the request is made; for the data to be processed, the merchant can act on the data from the Alipay feedback. The business logic it has to process the data (say orders) Now new changes, etc.). The results of the synchronization notification can only be used to present the results, and the inbound data should be based on asynchronous notifications.

3 Face recognition of brush face payment system

Face recognition technology is currently a popular technology. In fact, the system detects the face and extracts the feature information, and then queries relevant information in the system to feedback the identified identity. Recognition and detection will use image processing recognition, computer vision nerve acceptance and human brain analysis techniques and other related subject knowledge. Face recognition technology is currently widely used, for example, for credit card, record information, video telephony and the like. Because the face recognition system has the characteristics of not being in direct contact and fast and efficient, it has almost replaced the traditional identity verification system and applied to criminal investigation systems, entry and exit management, access control and attendance systems.

3.1 Face recognition process for payment systems

Face Detection Positioning: The system recognizes whether a face is input and accurately locates the face.

Face image preprocessing: can be summarized as normalization processing, specifically divided into geometry and grayscale processing. Geometric normalization means that the machine adjusts the recognized face to the same position and size to facilitate the next step of extracting information; gray-scale normalization is to perform image compensation such as illumination compensation and filtering. Because environmental factors such as on-site illumination or noise can cause interference to the machine recognition face, it is necessary to perform normalized pre-processing on the image to make the image light and dark and the size is adjusted to the same level.

Face feature extraction: Different face features are different, so the extraction method is different, but the feature extraction method is generally to obtain the characteristics of “small but unique” on each person’s face. When using the statistical recognition algorithm, the feature points of the classification are composed of the feature vectors and points of the portrait correlation matrix; however, when the geometric algorithm is used, the feature points are constructed by using the detected feature points.

Face recognition output: Compare the detected face with the face already stored in the system, and then feedback the results that have been queried or not queried to the page.

3.2 Face recognition algorithm based on convolutional neural network

$X^{(l, k)} = f (\sum_{p = 1}^{n_{i} - 1} (W^{(l, k, p)} \otimes X^{(l - 1, p)}) + b^{(l, k)})$ (1)

Where: X^(l,k) represents the k-th feature map of the l layer output, n_i represents the number of layers of the l-layer feature map, and W^(l,k,p) represents the p-th feature map of the l - 1 layer to the k-th feature map of the l layer The filter used when mapping the map. The generation of each set of feature maps of the l layer requires one filter and one offset. Assuming that the size of the filter is h × w, the number of parameters of the layer l convolutional layer is n_l-1 × n_l × h × w + n_l.

Common pooling operations have maximum pooling, averaging pooling, etc. The convolutional neural network in this paper uses only the maximum pooling. After the pooling operation, the size of the feature map will be reduced to the original 1/step according to the step size. The general form of maximum pooling can be expressed as follows. $E^{N} = \frac{1}{2 N} \sum_{i = 1}^{N} ∥ y_{i} - d_{i} ∥_{2}^{2}$ (2)

Where: N is the number of nodes in the neural network output layer, y is the predicted value of the neural network, and d is the manually labeled value. In this paper, we use the random gradient descent [6] algorithm to update all the connection weights and offsets between neurons. The final loss function is expressed as: $E = \frac{1}{m} \sum_{j = 1}^{m} (\frac{1}{2 N} \sum_{i = 1}^{N} {∥ y_{i}^{j} - d_{i}^{j} ∥}_{2}^{2}) + \frac{1}{2} η | ∥ W ∥_{2}^{2}$ (3)

Where: m is the number of training samples used for each backpropagation, and W is the weight matrix for each layer in the network. The weight matrix W in the convolutional neural network is updated during backpropagation. The network is initialized with a random value matrix W₀ before network training begins, and the weight matrix W_t+1 updated after t + 1 iterations can be expressed as follows. $W_{t + 1} = W_{t} - λ \cdot \frac{\partial E}{\partial W_{t}}$ (4)

3.3 Face feature point calibration algorithm

Facial features include eyes, nose, mouth, eyebrows and borders. These features can be used not only to analyze facial expressions, age, beauty index, but also as an important criterion for identifying the identity of a person’s face, so accurate feature points the positioning result becomes a prerequisite for extracting facial features. The mainstream feature point location algorithms include: 5-point, 25-point, 83-point, and so on. The basic flow of the feature point calibration algorithm includes: firstly, performing face detection on the input image to obtain an accurate face position; secondly, performing feature point detection according to the detailed texture information in the face frame; finally, assigning a specified number of feature points Calibration is in the face area. At present, both Uchi and Face++provide a cloud technology experience platform for enthusiasts to learn.

The structure of the convolutional layer and the down sampling layer alternately arranging not only the local and specific features of the front layer can be merged into a more global and abstract structural feature expression, which is beneficial to classification, and also simplifies the network structure in the process of feature dimensionality reduction. Eliminate interference from redundant information. The end of the convolutional network is usually the fully connected and output layers [7]. Among them, the fully connected layer rearranges the two-dimensional feature map into a one-dimensional feature vector in a certain way, and the output layer implements classification and function approximation. In the feature point calibration problem, since we want to obtain the coordinates of the five feature points, the number of nodes in the network output layer is 10, which represent the horizontal and vertical coordinate values of the five output points. The specific network structure is shown in Fig. 4.

In Fig. 4, it is assumed that the coordinates of the upper left corner and the lower right corner of the face frame detected by the face detection algorithm are (x₁, y₁) and (x₂, y₂), respectively, and the actual coordinates of the center point of the left eye pupil are (x, y), and the output of the output layer node is normalized. The value (a, b) is calculated as follows.

Fig.1

Alipay system platform architecture.

Fig.2

Data interaction mode.

Fig.3

Face recognition process.

Fig.4

For feature point calibration structure.

${\begin{matrix} a = \frac{{x - x}_{1}}{x_{2} {- x}_{1}} \\ b = \frac{y - y_{1}}{y_{2} - y_{1}} \end{matrix}$ (5)

The local weight sharing strategy is to divide the input feature map into p × q and other regions. The shared weight kernel weight is shared within each region, which is equivalent to extracting a local texture feature [8]. The schematic diagram of the local weight sharing strategy is shown in Fig. 5.

Fig.5

Local weight sharing strategy.

The local weight sharing strategy divides the original image into equal-sized regions, shares weights in each region, and extracts a texture feature of the region. Since the structural features inside each sub-area are relatively simple, the extracted features are more representative. Of course, the number of sub-areas can be selected according to the needs, and the number of sub-areas is too large, which increases the amount of calculation and leads to over-fitting. Therefore, it is especially important to select the number of sub-areas suitable for detecting objects.

4 Research on artificial intelligence face payment system

4.1 Environment construction

The application for the development of the Alipay platform is generally based on Google’s official website to provide developers with interface functions, software frameworks, hardware platforms, software development kits (SDK) and local development tools for C/C++(NDK), and complete code programming and integration of different components on Eclipse, and can be debugged with virtual system or real Android hardware device. Development system environment [9]: Windows 8.1. JDK: Java SE 8u51. SDK. NDK. Android SDK 824.3. 3-win64.Android NDK R10d. R&D environment: Eclipse 4.4.2 IDE.

4.2 System design objectives

(1) Design a user interface on the client side, and call the system camera interface to realize the face information acquisition, encryption and upload. (2) On the server side, the encrypted information is received and decrypted from the client, and then the face recognition technology is used to realize the location of the face and the identification of the identity information, and return the identification result and related rights. (3) Design a reliable information interaction path and encryption component to ensure the security and smooth interaction between the client and the server.

4.3 Overall structure of the system

The Android platform in this paper is a platform that can support the identity payment completion behavior through the face recognition system. The system module, the face information collection module, the face recognition module and other modules form the identity recognition and transaction realization functions. Figure 6 shows. Face acquisition module: Based on the Android platform, the front camera is used to collect relevant face information, and the collected information is encrypted, and this module is used in the registration and payment process. Communication module: Establish a reliable network information interaction path between the client side and the server side of the Android platform by mobilizing the Socket function. The face recognition module: completes the decryption of the received information, determines the identity through a series of processes such as human face information location, feature extraction and database retrieval similar information, and encrypts the search result back to the payment module. Payment module: decrypt the result returned by the server, and approve or cancel the payment according to the returned result [10].

Fig.6

Face recognition consumption.

Table 1

Analysis of some functional components of Alipay

Project	Description
Face recognition	1: N mode face recognition
	1:1 mode face verification Recognition rate > 99% (@FAR–0.1%)
	Recognition speed: 1 second
Report	If the identification is not passed three times, it can be reported and stored.
Personnel management	Increase user
	Query modification to delete user information
	Clear information
Port setting	Communication port setting
	Wigend settings
	Network settings
Face algorithm setting	Security level setting
	Exposure setting

4.4 System formation and testing

Registration and consumption provide two buttons for the user. The registration only needs to be selected when using for the first time. The face collection information uploaded to the server is stored in the known database, and the consumption button is selected when payment is required. Because the JSON format information is friendly to developers writing and reading, and easy to encode and parse information, a unified JSON (Java Script Object Notation) format protocol is used in this application design for data exchange. When the application is registered and consumed, it is represented by “opt” and assigned the corresponding Value. The value at the time of registration is “0000” and the value at the time of consumption is “1000”, the registered user name is “username”, and the payment amount is the character “payment”. The returned result is represented by the character “Remsg”. If the registration button is selected and the name tom is filled in the user name, an object structure is generated by the JSON format, “opt”.“10000” “username”: “tom”, after the server receives the encrypted information. Decoding is performed, and then the data is JSON parsed, and the identification of the value of the character “opt” determines whether the information from the client side is a registration request or a consumption request.

Collecting face image information mainly calls the front camera of the handheld device. Due to the openness of the Android platform, all applications provide interfaces to other applications and services when calling other applications and interfaces of services through the same framework and rules. The system directly calls the front camera hardware device through the Intent interface function to complete the photo shooting process and the image storage process. At the same time, while calling the front camera, through the face location function provided in the latest version of the Android platform, it is ensured that the captured and saved picture contains the facial information of the person, thereby avoiding the meaningless information uploading. In the latest Android platform API function interface, there is a special class method droid media Face Detector to achieve fast positioning of the face. The position coordinates of the eyes with symmetry in the image face are obtained by the calculation process to obtain the relevant face information, thereby providing parameters for the clipping of the face and the geometric normalization [11 –14].

The information exchange between the client and the Serve: uses the Socket information exchange based on the TCP link. The TCP-oriented link model mode ensures the order and integrity of information transmission. By activating a Sever Socket on the server side and calling the accept () method to listen to the client-side link request, the client activates a Socket and issues a request to link with the server. The server responds to the client’s request to complete the link construction, and then the client and server call the functions get Input Stream () and get Output Stream () to exchange information. Start a separate thread on the client side and the server side to avoid conflicts of information flow. The information stream is compiled and parsed in JSON format, and the image information transmission method is consistent with the file information transmission method.

The server returns the corresponding information through the day after tomorrow. If the registration option is selected, the registration success or failure is returned. The information format is “opt”: “0000”, “Remsg”: “registration succeeded!” If you choose the consumption option, the returned information is the identification result of the identity information.

5 Conclusion

In this paper, the system realizes the face information collected by Alipay platform and uploads it after encryption. The background face recognition receives the face information for decryption, then performs face detection and recognition, and returns the encrypted identification result. The results show that the proposed system has better recognition accuracy and system stability, but the collection of face images is easily affected by the outside world and needs further research and improvement.

References

, Chen

and Wang

, Face recognition attendance system based on pca approach, Caddm 26(2) (2016), 8–12.

Huang

S.M.

, Chou

Y.T.

and Yang

J.F.

, Low-resolution face recognition in uses of multiple-size discrete cosine transforms and selective gaussian mixture models, Computer Vision Iet 8(5) (2014), 382–390.

Paul

and Rho

, Probabilistic model for M2M in IoT networking and communication, Telecommunication Systems 62(1) (2016), 59–66.

Thakare Girish

, Prashant Dr. and Deshmukh

, Performance Analysis of Real Time Task Scheduling Algorithm, International Journal of Innovative Research in Computer and Communication Engineering (IJIRCCE) 5(6), 2017.

Tan

, Cao

and Cai

M.C.M.

, A new system of face recognition: Using fuzziness and sparsity. International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems 23(06) (2015), 829–844.

Bhattacharjee

, Paul

, Kim

J.H.

and Karthigaikumar

, An immersive learning model using evolutionary learning, Computers & Electrical Engineering 65 (2018), pp. 236–249.

, Face recognition with image sets using locally grassmannian discriminant analysis, IEEE Transactions on Circuits & Systems for Video Technology 24(9) (2014), 1461–1474.

Shubhangi T. Raut, Dr. Abha Khandelwal and Dr. S.B. Kishor, A Survey on Literature Review of the Usability of HCI, International Journal of Innovative Research in Computer and Communication Engineering (IJIRCCE) 5(7), 2017.

Bhatt

H.S.

, Singh

and Vatsa

, On recognizing faces in videos using clustering-based re-ranking and fusion, IEEE Transactions on Information Forensics & Security 9(7) (2014), 1056–1068.

10.

Singh

, Ganatra

, Shingrakhia

, Patel

and Motor

DC.

, Speed Control with PID Controller: A Review, International Journal of Innovative Research in Computer and Communication Engineering (IJIRCCE) 5(5), 2017.

11.

Farag

A.A.

, Bhanu

, Hancock

and Medioni

, Guest editorial special issue on facial biometrics in the wild, Information Forensics & Security IEEE Transactions on 9(12) (2017), 2019–2023.

12.

Coşkun

, Ayşegül Uçar, Özal Yildirim and Yakup Demir, Face recognition based on convolutional neural network, Modern Electrical and Energy Systems (MEES), International Conference on 2017.

13.

Samek

A.S.

and Sheremeta

R.M.

, Recognizing contributors: An experiment on public goods, Experimental Economics 17(4) (2014), 673–690.

14.

Moon

B.J.

, Smith

J.S.

, Ames

C.P.

, Shaffrey

C.I.

, Lafage

, Schwab

, et al., Prevalence and type of cervical deformities among adults with parkinson’s disease: A cross-sectional study, J Neurosurg Spine 24(4) (2016), 527–534.