BioPay: A Secure Payment Gateway through Biometrics

Due to emerging technological developments, major enhancements are taking place in the area of a secure and quick transaction. BioPay being a secure payment method is a one-step ahead. In the proposed methodology, there is no involvement of any credit or debit card or any other account information like OTP or CVV; it solely depends upon some unique identifying characteristic of a human known as biometrics. This work proposes a novel method that allows users to complete transactions quickly and securely using face and finger recognition. The transaction initiates with scanning face features and matching it with the database which in turn retrieves all the information associated with that customer account. After that, the system will scan the fingerprints of the subject and verify the transaction. This methodology can be implemented in ATMs and smartphones resulting in enhanced security and flexibility for payment purposes.


Introduction
Among the panoply of payment methods available today, credit cards and debit cards are the most used methods. People use cards for online shopping, ticket booking, money transferring, and ATM transactions. As every technological advancement improves the current state, there are also some pros and cons attached to every novel technology and invention so is the case with e-payment methods that offer ease of transactions on one hand while poses several challenges like vulnerability to suspicious activities and attacks on the other hand. People encounter various instances in their everyday lives where cards can be lost, stolen, or theft. Considering these insecurity concerns, the proposed work utilizes biometric to circumvent them, which enhances transaction security. The only pre-requisite is the user's face and fingerprint must be linked with the concerned bank.

Fingerprint Recognition
There available several methods for Fingerprint Recognition in the computer vision, such as correlationbased matching, minutiae-based matching, and pattern-based (or image-based) matching [19,22]. In the proposed system, Minutiae-based Fingerprint Recognition [1] is utilized.

Minutiae-Based Fingerprint Recognition
A preprocessed (Noise cleared, Binarised, and thinned) fingerprint image is fed to Feature Extractor which scans the local neighborhood of each pixel in the thinned image and computes the Crossing Number (CN) using: Every crossing number corresponds to a different ridge structure resulting in the extraction of different sets of minutiae points. For each Minutiae point, there will be (x, y) coordinate, Orientation factor ( ) type factor (CN). The extracted set is shown below: Then this image will be passed to minutiae matcher which will iterate over every minutiae point and find the closest image-based orientation and distance difference between minutiae point in the database fingerprint image and test minutiae image.
For orientation computation: Based on these observations the matrix MarkScore will be marked and the Match score with the formula: ℎ = 1 * 100 %

Face Recognition
For face as well, several methods are available and after a comparative study of all these different face recognition techniques in the cited paper [6,20,21,23,24], LBPH is found to be having better results than other techniques like PCA and LDA. In the proposed system, the LBPH classifier, a feature-based face recognition technology is utilized.

Local Binary Pattern Histogram (LBPH)
(LBPH) is a type of visual descriptor that works on grayscale images to recognize facial features. LBP is an ordered dataset ( ! , ! ) made with the comparisons of neighboring intensities of eight pixels with the central pixel intensity and given as: Where " corresponds to the intensity of neighboring pixel, # denotes the intensity of central pixel and is the threshold function and given as: The first step by the LBP classifier is to create a compressed image that can represent the original image, the LBP algorithm uses the concept of sliding with neighbor and radius as parameters to create the central compressed image. After that, the compressed image is divided into blocks, each block represents a histogram and aggregating all the histograms create a larger histogram, which can represent the original image. Now, This histogram will be compared with one in the database, and distance difference is determined using Euclidean distance (D) that can be used to determine the confidence level of the model.

Literature Sirvey
Biometric is becoming beloved and ubiquitous day by day, many kinds of research are going on enhancements of biometrics. One field is e-payment systems require extra endeavors, so embedding biometrics with payment methods for authentication can be a better practice. Many researchers investigating methods for authentication utilizing biometric. N. Badovinac et al. [2] proposed a multimodal biometric authentication system, that generates the required pin for card payments with the help of biometrics, without any involvement of the bank. Also, they have mentioned that the distance between the eyes and pupil will determine whether the person uses 4 digit pin or 6 digit pin and the one finger can generate at most 2 digit pin. So, the whole pin will be generated with the correct combination of multiple fingers. Now suppose if the person is using a 6 digit pin and has a small face for which their algorithm determines the person uses 4 digit pin, so transaction cancels or by some means of an accident, the person got scratches on one of these fingers, it will cancel the transaction. Also enlightening the accuracy of their system, Suppose if a finger matches with a 90% confidence level with the database, then the pin generated will be wrong as the whole procedure depends on the 100% accuracy of biometrics which is practically impossible.
By investigating methods for fetching the features of face for face recognition, the paper starts with is facial recognition. In [10], Olszewska et al. proposed system stated that Biometric technology is the automatic method to identify a person which we can use as a key step for our system, but their process depends on the physiological characteristics of a person which can not differentiate the twins. So to overcome that, we have used the iris recognition system along with face recognition. Also considering the credit card frauds, M. Chavan et al. [4] proposed a method for credit card authentication using face recognition by local binary pattern algorithms. In their method, they divided the facial recognition process into two categories: processing before detection where face detection and the alignment take place and afterward recognition that occurs through feature extraction and matching, but this system is cornered within credit cards due to the low-security level. So our system overcomes this limitation by adding extra biometric features like fingerprint recognition. With this biometrics combination, this system can push its limits from only use with credit cards to use in every payment that does not matter whether it's on a mobile phone, ATMs, or at shopping windows.
As the panoply of techniques available today for Biometric recognition and various researches are going on these techniques to improve their accuracy and success rate. Some of the cited papers dealing with these kinds of researchers are [6,9,15,18,24].

Architecture
This method requires a camera for face recognition and a fingerprint scanner. As there are no specific prerequisites, so there is no special architecture for the proposed payment method. This method can be utilized in smartphones and ATM's.

Methodology:
To develop and use our BioPay system, the steps required to be followed are given below: • Registration of the face and fingerprint data with the bank account Our proposed system comprises two main components, face recognition and fingerprint authentication that is required to be registered in the bank one time, so that registered data can be used as reference data during account fetching and fingerprint verification.

• Face Recognition
The first phase of the system is face recognition, which extracts the facial features like nodal points, retina structure, and iris structure and matches data with the database reference data by feeding facial data to the trained model at the server-side (LBPH classifier) so that it fetches the bank accounts linked with the particular face.

• Fingerprint Recognition
The second phase of the system is fingerprint recognition, which requires fingerprint from the user, this fingerprint will be preprocessed and filtered and extract the minutiae points of the input fingerprint with the help of fingerprint processing algorithm, then it matches it with the database. If a match succeeds, then the transaction is carried further.
• Confirmation of transaction from the bank servers Every bank has its own rules, like maintaining some minimum balance, so before carrying the transaction the bank servers will check all the formalities for a successful transaction like amount must be less than the funds in the bank account, Minimum balance is maintained, etc. After that Servers allow the transaction.

Experimental Results
For facial recognition, after demonstrating the LBPH algorithm, below mentioned are the first 10 iterations and corresponding accuracies in Table 3 and overall accuracy of the facial recognition model in Table 4:    In the proposed system, image is captured utilizing a webcam which is considered as a low-quality camera, also due to improper light causing illumination effect degrades the accuracy of the model. So they can be overcome by using a high-quality camera and proper lighting.

6.Conclusion & Future Scope
The proposed system manages payments utilizing biometrics such as fingerprint and face. This system. It initially scans the user's face and fed it to the proposed algorithm which will extract facial features and matches it with the database. Then, the window will show all the Bank accounts linked with that particular person, then the user has to select the transactional account. The system will demand for the fingerprint of the client to verify the transaction. If all biometrics matches, the transaction would proceed. As we are utilizing biometrics, it's a highly protected way to do a transaction. Also, it does not require any information regarding the bank accounts like card number, PIN, CVV, or bank account number. In the proposed Average Accuracy 88.73 Maximum Accuracy 92.74