Enhanced Elman Spike Neural Network optimized with Glowworm Swarm Optimization for Authentication of Multiple Transaction using Finger Vein

Nowadays, Automated Teller Machines (ATM) are broadly used every one Hence, the security is some more need to improve the bank sector. Due to increase in the count of criminals and their activities, ATMs have become unsafe. The access card and PIN are used in the ATM system for identity verification. Recent advances in biometric detection techniques, like fingerprinting, retinal scanning, face recognition, have made great strides in recovering the insecure environment at ATMs. In this manuscript, an Enhanced Elman Spike Neural Network Optimized with Glowworm Swarm Optimization is proposed for Authentication of Multiple Transaction Using Finger Vein (EESNN-GWO-AMT-FV). Here finger vein authentication, images are collected from the SDUMLA-HMT dataset. Then the images are pre-processed to improve the quality of the images using contrast limited adaptive histogram equalization filtering (CLAHEF).The features are extracted by using the visual geometry group network (VGG16).By using VGG16 model, various features are extracted, such as Vein Patterns, Local Binary Patterns, Dimensionality Reduction andImage Transformations. The extracted features are transferred to EESNN classifier for classifying the authorized person and unauthorized person. Then the weight parameters of the EESNN are optimized using the Glowworm swarm optimization Algorithm (GWO). The proposed method is implemented and the efficiency of the proposed EESNN-GWO-AMT-FV is examined under performance metrics, viz accuracy, specificity, sensitivity, precision, Error rate, AUC. The performance of the proposed method provides higher accuracy 99.01%, 98.34%, and 97.45%, and higher precision 87.12%, 94.12% and 91.78% compared with existing methods, like CNN-AOA-MBR-FV, CNN-MBR-FKP-FV and DCNN-MBR-FV respectively.


Introduction
Finger Vein authentication schemes commonly utilized in various applications for authentication purposes [1,2]. Unlike conventional authentication tools, like PIN, password always at risk of being forgotten or stolen, but the biometric authentication provides the best convenience for the user [3,4]. Privacy is an important concern in finger vein authentication systems [5]. If the biometric template is compromised, it cannot be changed. Several biometric template protection (BTP) schemes are presented to deal this challenge [6,7]. Even though BTP offers privacy for the finger vein authentication system, they suffer the performance of authentication [8,9]. Authentication depends on the vascular patterns formed by the blood vessels of the human finger in finger nerve biometric systems. Because these forms have sufficient characteristics, they are utilized for automatic personal recognition [10]. Most of these methods extract the structure of binary vessel, and then compare the extracted templates utilizing algorithms [11]. Besides these methods, some other methods use deep neural networks [12]. However, many works try to improve the finger vein recognition (FVR) systems secure. Index-of-Maximum (IoM) hashing technique has been used to present an alignment-free template protection scheme [13,14]. Conversely, in extracted biometric features, the user-specific random scheme is used to decrease the dimension of the features and create protected templates [15]. A deep neural network is applied secured templates.
Nowadays, crime in ATMs has become a matter of concern. There is no guarantee behind the security for the customer's account. Many who are unaware about PIN are not likely to memorize and recognize it, for eg, if they have lost their card, their account may be accessed by others and they may lose their money. To overcome this problem, some solutions need to be put forward to fix this problem. The existing methods did not provide sufficient accuracy for authentication of multiple transactions using finger vein, which are motivated to do this research work.  Here, finger vein authentication, imageries are collected from SDUMLA-HMT dataset [16].
 Then the imageries are pre-processed to improve the quality of the images using contrast limited adaptive histogram equalization filtering (CLAHEF) [17].
 Then features is extracted by using the visual geometry group network (VGG16) [18].
 By using VGG16 model, various features are extracted such as Vein Patterns, Local Binary Patterns, Dimensionality Reduction and Image Transformations.
 Then, the extracted features are transferred to enhanced Elman spike neural network (EESNN) [19] classifier for classifying the authorized person and unauthorized person.Then the weight parameters of the EESNN are optimized using the Glowworm swarm optimization Algorithm (GWO) [20].
 The proposed method is implemented and the efficiency of the proposed EESNN-GWO-AMT-FV is examined under performance metrics, like accuracy, specificity, sensitivity, precision, Error rate, AUC.
Remaining manuscript is organized as: section 2 portrays recent studies, section 3 describes about the proposed method, section 4 shows the results with discussion, and section 5 concludes the manuscript.

Literature Survey
Among the several research works related to authentication of multiple transactions using finger vein, some of the latest investigations are reviewed here, In 2020, Alay, et.al., [21] have presented deep learning approach for multimodal biometric recognition system based on fusion of iris, face, finger vein traits. The presented scheme was depending on CNNs that extract the features as well as classify the imageries through softmax classifier. To build the convolutional neural networks model, VGG-16, Adam optimization was deemed, also categorical cross-entropy was employed into loss function. It provides lower accuracy with minimum error rate.
In 2020, Daas,et.al.,[22]  In 2020, Shaheed, et.al., [24] have presented finger-vein presentation attack detection using depth wise separable convolution neural network. A depthwise separable convolution neural network (DSC) with residual connection and a linear support vector machine (LSVM) for automatic identification of finger vein presentation attacks. DSC was extract robust features from FV images. Then LSVM classifier was classify the images as bonafide and fake images. It provides low error rate with low accuracy.
In 2020, Yang, et.al., [25] have presented an embedded finger-vein recognition along antispoofing utilizing unified convolutional neural network. The finger-vein recognition and antispoofing network (FVRAS-Net) incorporates the recognition task and the antispoofing task into a unified CNN model utilizing multiple task learning model and attains higher security. A multiple intensity illumination was suggested into the embedded biometric system to automated select the most informative image for finger-vein identification, which enhances the recognition performance of the real system. It provides lower matching rate with higher specificity.
In 2021, Shahreza, et.al., [26] have presented protecting and enhancing vascular biometric recognition methods via biohashing and deep neural networks. Where, considered raw and pre-processed finger vein imageries and presented a deep-learning-base model for securing biometric templates as well as upgrade recognition performance. A deep convolutional autoencoder structure was used to lessen the feature space dimension, then secure templates in term of Biohashing approach. It provides lower error rate with lower precision.
In 2022, Heidari, et.al., [27] have presented deep learning based biometric authentication depending on different level fusion of finger knuckle print and fingernail. A multimodal biometric was considered to enhance the performance of authentication and make it resistant for spoofing attacks. A deep learning-based method with the help of convolutional neural network along AlexNet as a pre-trained model was applied. Various features were extracted from hand imageries, were consolidated normalization and fusion methods. It provides higher sensitivity with higher error rate.

Proposed Methodology
Cardholder can be verified by using fingerprint authentication as a tool to authenticate users at ATMs. ATM login process based on finger vein, which equipped with finger vein recognition technology can identify human finger vein during transaction. ATMs automatically remind the cardholder to be alert when there are "shoulder surfers" who try to peak over the cardholder's shoulder when the cardholder enters the PIN. The use of finger vein authentication enables to offer advanced transactions at its ATMs because finger vein authentication provides additional security to the cardholder and the bank. Bank customers are not limited to the accounts associated with their ATM card; they can access their accounts at the ATM using their debit / credit card. With finger vein authorization, the Bank can assure the most stringent security with the convenience of open banking service to its customers. This procedure utilized by the bank is to capture the customers face imagery at the branch of bank, then save the images in a secured biometric database. The deep finger vein activates the biometric software provided by the deep learning model when the customer taps his ID card or inserts his bank card in the ATM. The system captures multi finger vein imageries as well as find out the best image to be utilized for recognition. If nothing is deemed appropriate, the customer will be motivated to approach closer, or do whatever is necessary to capture the appropriate image. Figure 1 shows the Block diagram for EESNN-GWO-AMT-FV method.

Image acquisition
Here, the datas are gathered via SDUMLA-HMT dataset. This dataset comprise 3816 finger vein imageries that are captured via the device. This is generated by Intelligent Computing and Intelligent Systems of Wuhan University.

Pre-processing using contrast limited adaptive histogram equalization filtering (CLAHEF)
CLAHEF technique is utilized for improving the image quality and remove noise from input finger vein imageries. It has the ability to filter the small areas of the images and provides good outcome by removing noises. By using CLAHE technique, the images are preprocessed using equation (1) In this way, the input image noises are removed and the contrasts of the images are enhanced.

3.3.Feature extraction using visual geometry group network (VGG16) method
In feature extraction stage, pre-processed images are extracted using Visual geometry group network (VGG16). Visual geometry group network (VGG16) is one of the types of CNN model. To attain the feature map, convolutional layers are utilized for convolving the preprocessed image by means of kernel weight. The kernel weight is connected with the feature map units of preceding layer. VGG16 contains twenty three layers with weights. The initial layer is convolutional layer used for pre-processing and it reduces the preprocessed image size by removing soft tissues. Next, the pre-processed image size is decreased to . Thus, the feature extraction using convolution layer equation is given in equation (3).
Where, h is represented as the output of convolution layer,W , a are represented as preprocessed inputs from the previous layer with weights, d f , are represented as the size of the convolution matrix, i j, are represented as th i and th j finger vein image from database MD ,  is represented as the activation function. Then, the output of this layer that is layers 2 and 6 apply activation function from attained convolutional output h . The next layer is ReLU layer, that is, layer 3 and layer 7 is linear activation function to the neuron output is expressed in equation (5), Next, the layer four and eight are the cross-channel Norm layer with five channels. Layer five and layer nine contains 3 3 max-pooling layer. Here, max-pooling layer is utilized for down sampling process that lessens the over fitting problem through the classification method. By Vein Patterns are extracted through these images categories. In several cases, topological or curvature information of the veins on the basis of preprocessing steps for higher degree, because the veins are in the binary image and better performance of every method. The feature extracted from vein patterns are given in equation (6)  Local Binary Patterns (LBP) extracting the region of interest for finger vein, and extract the LBP images. As an alternative, the histogram image is considered for matching process at certain cases. LBP is calculated by utilizing the pixel intensity along neighbor's input preprocessed image and it is expressed in equation (7).
Where, i , j specifies the input pre-processed image,  

Classification using Enhanced Elman Spike Neural Network (EESNN)
Here, Enhanced Elman Spike Neural Network (EESNN)is utilized for classifying authorized and unauthorized person. The EESNN is modified by basic Elman NN and this is one hind of partial recurrent spike neural network model. This EESNN has 4 layers, such as input, context, invisible, output. The modified structure contains self-feedback variable that gain in the context layer and the neural network is fed with the input ) The basic function and propagation of every layer is given to authentication purpose. The input layer with node layer is shown in equation (10),  (11), is denoted as the sigmoid function of authentication and it is shown in Then, the context layer is shown in equation (13) ) Where,  represents self-connecting feedback gain contains 1 0    . Then, the output layer is shown in equation (14) as given below,  (17), Then, the weight jk T is adjusted by equation (18), Where 1  represents the detection layer and it is updated using equation (19), Where 1  represents the training rate of layer four. Update the weight parameters of invisible layer in lj T , then the training pattern of weight lj T is adjusted using equation (20) ) Next, the connecting weight lj T is updated using equation (21) lj lj lj The attraction of glowworm is made by its sparkling neighbor within its vector space. The step-by-step process of GWO are described below,

Step 1: Initialization
Initially, all the glowworms carry an equal luciferin level randomly based on the lower and upper bounds of the production power of glowworm and control parameters. The initial population of glowworm is expressed in equation (22), Where, s Q and d denotes the lower and upper bound of the parameter.

Step 2: Random Generation
After the process of initialization, the input parameters are generated randomly. Here, the highest fitness values are selected depends on accurate hyper-parameter context. Generate randomly the population of procedure value for the accurate prediction of authentication in multiple transactions.

Step 3: Fitness Function
This is assessed to reach the objective function that is accurate prediction of authentication in multiple transactions as well as reach the optimum value. The weight parameters of EESNN , Tansig represents the tangent sigmoid functions, g indicates the number of the current iteration and max g is the maximum number of iteration.
Step 6: Update luciferin volume to optimize the weight parameter jk T The luciferin volume is updated to optimize the weight parameter jk T . Exploration of glowworm for better solutions are obtained by equation (25), Where,  denotes the randomly selected position j for glowworm i and i j NI represents the new source. Figure 2 shows the flowchartfor Glowworm Swarm Optimization Algorithm for optimizing EESNN.
Step 7: Perform mutation operation to optimize the weight parameter * s  In GWO, the mutation operation performed is with respect to probability values using the fitness values provided by glowworm. To this intention, a fitness based selection technique is used. It is obtained using equation (26), Where, the training data of EESNN for classifying the authorized and unauthorized person expressed as ) (M N , g expresses the current iteration number, moreover max g denotes the best optimal locations, S indicates the maximum number of iterations and R denotes the round. Certification and authentication processes for multimedia messaging services (MMS) require robust Internet and GSM networks. Many anti-fraud measures are structured to add to its security in the system. The infrared lens is to capture additional facial details to prevent fraudulent attempts.

Result and Discussion
In this section, the experimental result is discussed for Enhanced Elman Spike Neural

Dataset description
Here, the datas are gathered via SDUMLA-HMT dataset. This is a multimode biometrics images. Every image is stored in "bmp" format with 320×240 pixels in size. The total size of finger vein database is around 0.85G Bytes.

Performance measures
This section describes the performance measures needed for experiment. The performance metrics is a significant role for the authentication of multiple transactions using finger vein.
To authenticate the performance, the most common performance measures, like accuracy,

Accuracy
This is computed by,

Specificity
This is computed by,

Sensitivity
This is nothing but recall or detection rate, it is computed utilizing the given equation,

Precision
This is determined by,     Figure 9 depicts Screen for getting input from user. After entering the vein, it processed and a message box will be displayed, which intimate that the authentication is passed. Figure 10 depicts Screen displaying whether the authentication is passed or not. Then the user is allowed to access his/her multiple account by entering the two digit pin number using the keypad. Otherwise, message box with an authorized user is displayed, if the user's vein is not matched.

Conclusion
An

Data availability statement
Data sharing is not applicable to this article as no new data were created or analyzed in this study.

Funding Information:
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Ethical approval:
This article does not contain any studies with human participants performed by any of the authors.

Declaration of interest Statement
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper