2024 2nd International Conference on Artificial Intelligence and Machine Learning Applications Theme: Healthcare and Internet of Things

(AIMLA) | 979-8-3503-4922-1/24/$31.00 ©2024 IEEE | DOI: 10.1109/AIMLA59606.2024.10531351

2024 International Conference on Artificial Intelligence and Machine Learning

Applications(ICAIMLA), March 15 – 16, 2024, Tiruchengode, India

Hand -Based Multimodal Biometric

Authentication System
Baratam Siddardha S.Veluchamy,
Department of CSE (Cyber Security) Assistant Professor, Department of CSE (Cyber Security)
Amrita Vishwa Vidyapeetham, Amrita Vishwa Vidyapeetham,
Amrita School of Computing, Chennai, Tamilnadu, India Amrita School of Computing, Chennai, Tamilnadu, India
[email protected] [email protected]

This Project initiative takes on a creative journey to overcome

Abstract— Biometric authentication technologies have become these constraints by combining the algorithms of finger’s vein
more well-known recently because of their high accuracy and and knuckle recognition systems in response to these
security in validating individuals' identities. Finger’s vein and challenges. The goal of this fusion project is to open up a world
knuckle biometrics are two of the most promising biometric of enhanced accuracy, strengthened security, and unmatched
modalities for personal identification. By combining the dependability in personal identity. By integrating multiple
distinctive characteristics of both finger’s vein and knuckle biometric features, using numerous aspects within a
biometric recognition practices, this technology offers a
multimodal approach improves security and dependability by
revolutionary way to improve biometric security and accuracy.
Finger-knuckle recognition is concerned with the shape and increasing the difficulty of spoofing or unauthorized access.
creases of the knuckles, whereas finger-vein technology examines Through leveraging the natural complementarities between
the distinct vascular patterns found within each finger. The these two unique biometric modalities, this study aims to
system overcomes individual constraints and improves security develop an authentication system that is robust enough to
and accuracy by integrating these two methods. A fusion function well in practical situations. An important aspect of the
algorithm and an extensive dataset are developed by the proposed research is developing a sophisticated fusion method to
system for training and testing. This system may be used for
access control and identity verification, as well as to increase seamlessly merge feature vectors from finger-vein and finger-
security and decrease false acceptances. By advancing biometric knuckle pictures. Finger-vein and finger-knuckle biometrics
authentication methods, this research promises a safer digital together constitute not just a research endeavor but also a
future. significant development in safe and dependable identification
methods. Along with potentially solving security problems
Keywords— Finger-Knuckle Recognition, Vascular patterns, related to identity verification and access management, it
Fusion algorithm promises to further the field of biometric authentication
technology.
I. INTRODUCTION
II. RELATED WORKS
In this modern era, where digital identity and safe access are Hengyi Ren [1] developed a novel method for
required, there is an increasing demand for creative and encrypting finger vein images that leverages the RSA
dependable ways of personal identification. Biometric encryption mechanism. A completely reversible finger vein
technologies have emerged as pioneers in this developing field, detection system with template safeguarding is also
capitalizing on people's inherent, distinctive characteristics. recommended to maintain recognition performance and ensure
Finger’s vein and knuckle biometrics are two biometric the privacy of the user's vein template. Four public databases
modalities that have gained popularity; each has advantages and have been the subject of a thorough series of comparison tests
disadvantages of its own. While finger-knuckle recognition and security evaluations.
investigates the fascinating realm of finger creases and Su Tang [2] analyzed the drawbacks of current finger
geometry, finger-vein recognition goes further, thoroughly vein verification algorithms that depend on manually created
investigating the complex vascular patterns unique to each features and suggested a novel method that makes use of CNNs,
person's finger. Despite their potential, these biometric Siamese structures, and knowledge distillation. To ensure
approaches have certain drawbacks when it comes to identity reliable feature extraction in the context of finger vein
verification. While finger-vein recognition may be more verification, the authors stress the significance of overcoming
vulnerable to external stimuli, finger-knuckle recognition may obstacles associated with a lack of training data.
struggle with variations in finger location. Zhitao Wu [3] applied GCNs in fusion recognition and
the graph structure for feature extraction.

uthorized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on October 22,2024 at 02:46:01 UTC from IEEE Xplore. Restrictions apply
 Carlos Vivaracho‐Pascual [4] focuses on non- Marius Iulian Mihailescu [10] proposes a three-factor
probabilistic classifiers and expands on previous research by authentication approach which includes searchable encryption,
proposing Score Ratio as a substitute approach for decision- RSA-based passcode, and a token or password. It focuses an
making in biometric systems. Though the paper suggests that intense focus on the safe recovery of data from a suspicious
the Score Ratio can offer improvements in non-probabilistic cloud server via biometrics. The approach is included in the
scenarios, the similarity rate is still frequently utilized with overall structure of multi-factor authentication, accentuating
probabilistic classifiers. Utilizing databases, classifiers, the role that searchable encryption, biometrics, and RSA serve
performance metrics, third-party experimental protocols, and in the proposed approach.
reproducibility tests, the methodology is assessed. When the Satyendra Singh Thakur [11] exhibits that a wireless
Score Ratio is applied in the tested cases, the results show networking system (MIMO-OFDM) and finger vein pattern-
statistically significant improvements when compared to not based biometric authentication are incorporated. With the goal
applying it. of enhancing precision in recognition, it promotes the
Syed W. Shah [5] especially focuses on hand vein utilization of hybrid feature extraction and feature selection
patterns, adds to the expanding field of biometric identification approaches. For confidential communication, the Dual-RSA
systems. Infrared imaging, deep learning for biometric conduct is used, and for classification, an error-correcting
applications, and depth sensing are the areas where it expands output code-based SVM is used. In the structure of biometric
on previous research. The current trend of employing authentication and remote networks, the research aims to
sophisticated neural network architectures for reliable pattern address the challenges of feature optimization, encrypted
recognition in biometric authentication systems is reflected in communication, and accuracy improvements.
the integration of CNN and Stacked Autoencoders. Kevin H. M. Cheng [12] underlines the weaknesses in
Bernhard Prommegger [6] points out that, despite the current 3D finger-knuckle authentication approaches and the
field's current efforts, no method exists to address all forms of need for more efficient matching approaches. When compared
finger misplacements, with longitudinal finger rotation being a to cutting-edge methods, the suggested method significantly
significant issue. To increase recognition rates and reduce the improves speed (almost 23 times quicker) and accuracy by
need for multiple cameras, a multi-camera system that captures utilizing surface key points extraction. The work portrays the
vein patterns from various angles is the suggested solution. efficiency of the suggested approach across various datasets by
Farah Bahmed [7] offers a unique multimodal extending its validation to other 3D biometric structures, such
biometric system that combines the distal phalanx and as 3D fingerprints and 3D palmprints.
metacarpo, contributing to the field. Promising results on a Arya Krishnan [13] states that it is the first study to
standard database from experimental tests indicate that the use the pulsations of the veins in the fingers for biometric
ALLBP method is an improved feature extraction technique for identification, indicating a unique strategy in contrast to current
this region. image-based techniques.
Javad Khodadoust [8] demonstrates that the current Junying Zeng [14] identifies the drawbacks of the
multibiometric systems are not meeting the necessary demands current techniques for extracting features from finger veins and
on accuracy levels, the ideal number of biometric modalities, suggests a comprehensive model that combines FCN and CRF.
and economic concerns. It also points out that multibiometric It incorporates advances such as sliding window summation,
systems tend to focus primarily on verification mode, with deformable convolution network, and the integration of RNN
identification mode being a major difficulty. Filling these and ResNet to improve feature extraction. According to three
deficiencies, the suggested system introduces an identification publicly available finger vein datasets, the suggested model
mode algorithm and uses three modalities from a single finger, performs better on finger-vein verification tasks than
improving recognition accuracy in tests on benchmark comparable models like U-Net.
databases. Shuyi Li [15] introduces semantic and visual
Sharmila S. More [9] focuses on how cryptography is uniformity in feature representation to overcome the
becoming more and more important in computer security, aforementioned problems. Experiments on four popular finger-
especially when it comes to solving security-related problems vein databases show that our method works better than current
in a variety of technical sectors. AES is used to overcome the finger-vein recognition techniques.
constraints of the DES algorithm, and the use of biometrics with
multimodal—such as fingerprints, iris, faces, and palm prints— III. METHODOLOGY
is emphasized. In addition to discussing the difficulties with For this project, we have reviewed 20 publications in the early
matching algorithms, time delays, and FAR in multimodal stages, investigated the limitations of existing biometrics
biometrics, the literature review emphasizes the contribution identification models, and developed a solution—using a
that cryptographic methods provide to improving the security multimodal biometric system that combines knuckle and finger
of biometric data. Furthermore, the work provides a MATLAB vein identification—to get around those problems. Due to its
implementation of the AES algorithm, analyzing important inherent advantages of security, ease, and liveness detection,
variables including key size, input size, processing time, finger vein identification has emerged as a potential biometric
simulation, memory use, and CPU utilization. authentication method. This device reliably identifies users by
using their distinct vein patterns beneath the skin's surface.

uthorized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on October 22,2024 at 02:46:01 UTC from IEEE Xplore. Restrictions apply
 Accurate and efficient authentication is made possible by a Squeeze Net. It is appropriate for environments with limited
sequence of clearly defined steps in the system design. resources since it leverages fire modules to lower the number of
parameters without sacrificing performance. The Squeeze Net
architecture was implemented, emphasizing its computational
efficiency and lightweight design for knuckle feature extraction.
Made use of Squeeze Net's capability to retain high accuracy
while utilizing fewer parameters, which qualifies it for knuckle
image analysis.
E. Feature Fusion
Consolidated the feature sets taken from Squeeze Net
(knuckle) and Efficient Net (vein) into a single, all-
encompassing feature representation. Combining data from vein
and knuckle modalities was designed to enhance the precision
and durability of the biometric authentication system. The
complicated process of fusing the vein and knuckle features was
one major challenge. To provide accurate and dependable
authentication, a thorough understanding of how to process and
fuse the features was necessary for the integration of these two
modalities.
F. OpenSSL- AES Encryption
To provide robust cryptographic protection, the fused feature
set was encrypted using the Advanced Encryption Standard
(AES). AES encryption was implemented using the OpenSSL
Fig.1. Architecture Diagram
library, a well-known and open-source toolkit for secure
communication and data protection.
A. Data Collection
G. Testing
Gathered a variety of photos that show each person's
distinct vein patterns and knuckle characteristics. For To compute the performance of the proposed multimodal
biometric modality, extensive testing procedures were carried
reliable model training, it was made sure the gathered data
out. Evaluated the system's precision in user identification and
reflected a wide range of variances in vein and knuckle
properties. authentication using the fused features that Squeeze Net and
Efficient Net extracted. Examined the model's effectiveness in
B. Preprocessing terms of resource usage and processing speed. Verified the
Used methods like contrast enhancement and noise encrypted feature set's security to make sure it is resistant to
reduction for images. Image sizes and orientations were manipulation or unwanted access.
standardized to guarantee uniformity in later analysis. The proposed approach uses cutting-edge methods and
C. Feature Extraction (Vein) resources at every level of development, namely Efficient Net
for vein feature extraction, Squeeze Net for knuckle feature
Efficient Net is used in the field of deep transfer learning to extraction, and OpenSSL for AES encryption.
extract vein features. The Efficient Net class of convolutional
neural networks is unique because it balances the network's
depth, breadth, and resolution to provide cutting-edge accuracy IV. RESULTS AND DISCUSSIONS
and efficiency. This enables it to be computationally efficient
and attain great performance in a variety of activities. To extract For classification, the approach suggested works well for
vein features, the Efficient Net architecture was used, which encrypting and combining biometric features. The suggested
made use of its effectiveness and efficiency in capturing minute approach's excellent classification accuracy indicates that it has
details. Vein photos were automatically processed to obtain practical applications. An understanding of how well the
discriminative features using Efficient Net's deep neural suggested method works in various classes is given by the
network structure. confusion chart.
D. Feature Extraction (Knuckle)
Squeeze Net is used in the field of deep transfer learning to
extract knuckle features. The deep neural network with a
reputation for being compact and highly effective is called

uthorized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on October 22,2024 at 02:46:01 UTC from IEEE Xplore. Restrictions apply
 V. PERFORMANCE ANALYSIS
In this section, the suggested method's performance analysis
using several SVM kernels is discussed. This is divided into
two subsections. The first component contains the confusion
matrix values for each of the six SVM classifiers. The second
section's goal is to evaluate the recall, accuracy, and precision
values of several SVM classifiers to identify which produces
the best outcomes.

A. Performance of SVM Classifiers

The graph displays how various SVM classifiers performed
using a given dataset. True Positive -TP, False Positive - FP,
False Negative - FN, and True Negative - TN metrics are used
to assess the classifiers. The performance of the classifier
improves with increasing TP and TN values and improves with
decreasing FP and FN values.

Fig.2. Confusion Matrix

High classification accuracy is achieved by the proposed

approach for both the vein and knuckle datasets, according to
the confusion matrix. With a specific accuracy of 99.5% for the
knuckle as well as 96.5% for the vein, the total accuracy is 98%.
With only a small percentage of samples misclassified, the
confusion chart demonstrates that the model can accurately
classify most samples. Fig.4. Performance analysis of SVM kernels

B. Comparison of Precision, Accuracy, and Recall for

Different SVM Kernels
Three metrics—precision, recall, and accuracy—are used to
compare the effectiveness of several Support Vector Machine
(SVM) kernels on a dataset in the figure.
• Precision is the percentage of actual positive results among
all positive predictions. Here is the formula:

• Recall: The percentage of true positives that are actual

positives. Here is the formula:
Fig.3. Training data analysis

The analysis of the training data chart indicates that as the

number of training samples rises, the accuracy of the training • Accuracy: The percentage of samples that are accurately
data rises consistently. Approximately 500 training samples are classified. Here is the formula:
needed for the accuracy to reach a saturation point, indicating
that the model has picked up the most crucial features from the
instruction set. With a 98% overall accuracy rate for the training
set, the model is well-trained and should function well when The several SVM kernels that were employed are displayed on
faced with new data. the chart's x-axis: Linear SVM, Quadratic SVM, Cubic SVM,

uthorized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on October 22,2024 at 02:46:01 UTC from IEEE Xplore. Restrictions apply
 Fine Gaussian SVM, Medium Gaussian SVM, and Coarse VII. REFERENCES
Gaussian SVM. [1] Ren, Hengyi & Sun, Lijuan & Guo, Jian & Han, Chong & Wu, Fan.
The precision, recall, and accuracy values for each kernel are (2021). Finger vein recognition system with template protection based on
displayed on the y-axis. convolutional neural network. Knowledge-Based Systems. 227. 107159.
10.1016/j.knosys.2021.107159.
[2] Tang, S., Zhou, S., Kang, W., Wu, Q. and Deng, F. (2019), Finger vein
verification using a Siamese CNN. IET Biom., 8: 306-315
[3] Wu, Z., et al.: Robust graph fusion and recognition framework for
fingerprint and finger-vein. IET Biome. 12(1), 13–24 (2023)
[4] Vivaracho-Pascual, C., Simon-Hurtado, A. and Manso-Martinez, E.
(2021), Improving biometric recognition by means of score ratio, the
likelihood ratio for non-probabilistic classifiers. A benchmarking study.
IET Biome, 10: 127-141.
[5] Shah, S.W., Kanhere, S.S., Zhang, J. and Yao, L. (2021), VID: Human
identification through vein patterns captured from commodity depth
cameras. IET Biome, 10: 142-162.
[6] Prommegger, B. and Uhl, A. (2021), A fully rotation invariant multi-
camera finger vein recognition system. IET Biome, 10: 275-289.
[7] Bahmed, F. and Ould Mammar, M. (2021), Basic finger inner-knuckle
print: A new hand biometric modality. IET Biome, 10: 65-73.
[8] Khodadoust, Javad & Medina-Pérez, Miguel & Monroy, Raúl &
Khodadoust, Ali Mohammad & Mirkamali, Seyedsaeid. (2021). A
Fig.5. Comparison of Recall, Accuracy and Precision multibiometric system based on the fusion of fingerprint, finger-vein, and
finger-knuckle-print. Expert Systems with Applications. 176. 114687.
VI. CONCLUSION AND FUTURE SCOPE 10.1016/j.eswa.2021.114687.
[9] More, Sharmila & Narain, Profbhavana & Jadhav, Bharat. (2021).
To strengthen the safety of data, we utilized both vein Advanced Encryption Standard Algorithm in Multimodal Biometric
and knuckle features in this research to demonstrate a novel Image. 10.1007/978-981-15-6329-4_7.
approach to biometric authentication that uses encryption. [10] M. I. Mihailescu and S. Loredana Nita, "Three-Factor Authentication
Scheme Based on Searchable Encryption and Biometric Fingerprint,"
The proposed approach has been shown to achieve high 2020 13th International Conference on Communications (COMM),
classification accuracy for both knuckle and vein datasets, Bucharest, Romania, 2020, pp. 139-144, doi:
with an overall accuracy of 98%. The results illustrate the 10.1109/COMM48946.2020.9141956.
[11] Thakur, S.S., & Srivastava, R. (2019). Dual RSA Based Secure Biometric
productiveness of the suggested methodology in combining System for Finger Vein Recognition. Inventive Computation
and encrypting biometric features for classification. Technologies.
[12] K. H. M. Cheng and A. Kumar, "Efficient and Accurate 3D Finger
The proposed approach can be expanded in several ways. Knuckle Matching Using Surface Key Points," in IEEE Transactions on
To begin, the approach's performance can be evaluated on Image Processing, vol. 29, pp. 8903-8915, 2020, doi:
10.1109/TIP.2020.3021294
larger and more diverse datasets to ensure its generalizability.
[13] A. Krishnan, T. Thomas and D. Mishra, "Finger Vein Pulsation-Based
Secondly, various encryption algorithms can be examined to Biometric Recognition," in IEEE Transactions on Information Forensics
evaluate how they affect data security and classification and Security, vol. 16, pp. 5034-5044, 2021, doi:
accuracy. Third, to assess its practical applicability, the 10.1109/TIFS.2021.3122073
approach can be integrated into real-world applications such [14] J. Zeng et al., "Finger Vein Verification Algorithm Based on Fully
Convolutional Neural Network and Conditional Random Field," in IEEE
as access control systems and mobile banking platforms. Access, vol. 8, pp. 65402-65419, 2020, doi:
Furthermore, the impact of different machine learning 10.1109/ACCESS.2020.2984711.
models on classification performance can be investigated. [15] Liu, Haiying & Yang, Gongping & Yang, Lu & Yin, Yilong. (2019).
Learning Personalized Binary Codes for Finger Vein Recognition.
Neurocomputing. 365. 10.1016/j.neucom.2019.07.057.

uthorized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on October 22,2024 at 02:46:01 UTC from IEEE Xplore. Restrictions apply