Volume 7, Issue 8, August – 2022 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Neural Network-Based Fingerprint Matching and

Human Recognition System
Ene Princewill Chigozie
Electrical and Electronic Engineering Department,
Faculty of Engineering, Enugu State University of Science and Technology (ESUT)

Abstract:- This research presents a neural network- The use of technology has provided many solutions for
based fingerprint matching and human recognition payroll fraud using a biometric approach. Biometric
system. The methods used are data collection, data technology employed the use of physical and behavioral
processing, data extraction, and convolutional neural traits of human beings such as their eyes, face, handwritten,
network (CNN). The CNN was adopted and used to fingerprint, voice, and iris among others, for person
develop the human identification system based on object- identification [10; 11].These biometric traits all have their
oriented methodology and then implemented with advantages and disadvantages as identified in [12], but the
Simulink. The result when tested showed that the new use of fingerprint technology has been the most consistent,
system was able to verify humans with the fingerprint at easy to use and affordable, reliable, and most used biometric
90% accuracy. technology all over the world when compared to other
biometric counterparts.
Keywords:- fingerprint, Convolutional Neural Network,
human identification, accuracy. Fingerprint-based technology employed the use of
fingerprint data for human identification. Many works have
I. INTRODUCTION employed various techniques to develop this system such as
image processing, fuzzy logic, expert system, and machine
Every day begins a story and this time, the narrative learning among others, however, the use of machine
was inspired by the ever-increasing rate of fraud across the learning techniques has provided a better solution when
globe. Fraud is simply a criminal and unlawful act to makea compared to others [13; 14; 15].
financial gain. This has eaten deep into the public and
private sectors and has remained a major challenge. Fraud Machine learning according to [16] isa series of
can occur in many forms such as impersonation, payroll algorithms thatcan learn and do classification or regression
fraud, malpractices, fabrication of fake documents, etc [1]. problems. These algorithms include K-nearest neighbor,
support vector machine, clustering, and Artificial Neural
Today the common trend is the issue of payroll fraud Network (ANN), among others, however, [17] have
[2]. Payroll fraud is an act of stealing money from an identified the use of the artificial neural network as the most
organization by the ghost worker perpetrators. Ghost reliable in solving image-based pattern recognition
workers are employeeson a payroll, but do not work for the problems. [18; 19; 20] used an artificial neural network to
company [3]. This has remained a challenge all over the solve the problem of human identification, but despite its
world and has crippled many economies.[4] Reportedly in success, it is not considered a solutionto the problems of
Borno state Nigeria, 22,556 ghost workers were discovered payroll fraud. The solution to the aforementioned problems
within the state government payroll system [5], further was proposed in this work as a development of a neural
investigation revealed that the introduction of the Integrated network-based fraud detection system using fingerprint
Payroll and Personnel Information System (IPPIS) matching technology.
uncovered 60,450 ghost workers in the Nigerian government
payroll system, among other cases mentioned in [6; 7; 8], II. METHODOLOGY
however, the IPPIS did not address the peculiarity of the
Nigerian University education system of operation. The methodology used for this system development is
the object-oriented analysis methodology. The methods
To solve this problem,[9] presented some include data collection, data processing, feature extraction,
administrative solutions which include the use of classification, and results.
technology, transparency in the payroll system, and  Data collection: Data collation activity was done at the
accountability mechanisms, among others, however, the Enugu State University of Science and Technology
study revealed that the use of technology will provide a (ESUT).Fingerprint data of the students and staff were
more reliable solution to ghost worker fraud when compared collected using the KOJAK fingerprint scanner as shown
to the rest. in figure 1;

IJISRT22AUG639 www.ijisrt.com 1043

Volume 7, Issue 8, August – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Fig. 1:Setup for data acquisition

The setup was used to collect 515 fingerprint data of feature extraction techniques and concluded that it is the
students and nonacademic staff as the sample size. These best.
data were formatted using image quadratic programming-  Classification: This process used a feed-forward artificial
based image resizing techniques in [22] to the same size and neural network to learn the fingerprint data and then
then stored in the training dataset. generate a classification algorithm. The neural network is
biologically inspired neurons that have weights, bias,
 Data Processing: Fingerprint data always contain noise hidden layers, and activation functions to learn the patterns
that can originate from the nature of the query person's of the training data features and generate a reference
finger, dirty on the scanner, among other environmental verification model used for the verification of a person.
sources. This results in the poor quality of the data when  Result: This is the verification output of the neural
collected and hence the need for data processing. The data network when tested with a test fingerprint set.
was processed using a filter developed by [23]. The filter
was used to remove noise and enhance the features of the III. SYSTEM DEVELOPMENT
data for better classification results.
 Feature Extraction: This process was used to extract the The system development used a structural method that
fingerprint pixels into a compact feature vector. This was shows the logical interaction of the steps used to achieve the
achieved using a sequential binarization approach to new system. The development first configured neural
extract the image's interesting features using thinning network architecture as shown in figure 3; using the
algorithm adopted from [25]. The reason behind the attributes of the training dataset and then training the neuron
feature extraction choice was because [25] has with the data to learn and generate the desired classification
comparatively evaluated its performance with other reference model.

Fig. 2: The Artificial Neural Network Architecture

Figure 2 presents the structure of the neural network neurons are trained by loading the data into the neural
configured to train the fingerprint data.This was achieved network as a training and test set using the neural network
using the Rectified Linear Unit (ReLU) activation function toolbox to generate the configured neural network model in
which is the most used in developing neural network Figures 3a and 3b.
structures for pattern recognition problems [27]. The

IJISRT22AUG639 www.ijisrt.com 1044

Volume 7, Issue 8, August – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

(a)

(b)

Fig. 3: Simulink model of the configured neural network

IJISRT22AUG639 www.ijisrt.com 1045

Volume 7, Issue 8, August – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Figure 3(a) and 3(b) present the configured neural analyzer embedded in the neural network tool to check the
network when loaded with the fingerprint data for training. training steps at epoch intervals until the best result was
To train the neurons, the back-propagation algorithm and the achieved, indicating that the data were learned by the
ReLU were used to learn the feature vectors until the best neurons. The training parameters and pseudo code of the
training result was achieved. This was evaluated using fingerprint algorithm are presented in table 1;
epoch parameters, Mean Square Error, and a regression

Finger Training parameters Values Pseudo Code Of Fingerprint Verification System

Maximum number of epoch to train 15 1) Start
Epoch between display 5 2) Load fingerprint data
Maximum time to train in sec Infinity 3) Split data into train, test, and validation set
Maximum validation failure 5 4) Configure neural network architecture
Scale factor for length 12 5) Train data
Scale factor for weight 10 6) Initialize epoch values
Initial step size 0.01 7) Train data
Minimum performance gradient 1e-6 8) Test and validate the model
Cost horizon 7 9) If
Control horizon 2 10) Desired Mean square error and Regression is achieved
Number of bias function 1 11) Stop training
12) Generate fingerprint reference model
Number of input 2
13) Else
14) Adjust bias and weight function
15) Continues training
16) End if
17) End
Table 1: Training parameters of the neural network

IV. IMPLEMENTATION

The system was implemented using the neural network toolbox to load, configure and train the data as shown in figure 4;

Fig. 4: The Artificial Neural Network training tool

IJISRT22AUG639 www.ijisrt.com 1046

Volume 7, Issue 8, August – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Figure 4 presents the neural network tool in Simulink Figure 4 was used to evaluate the training performance until
used to train the data. When the data was uploaded into the the desired result was achieved. The training toolbox used to
tool, it was automatically divided into test and training sets evaluate the performance is presented in The (d) part of
in the ratio of 80:20 and then used to configure the neural Figure 4 and the results are discussed in the next section.
network as shown in figure 4 (a). The (b) part of Figure 4 The generated algorithm was deployed into Mathlab and
was used to train the neurons andthe backpropagation developed as a biometric person identification system as
method was used to learn the feature patterns.The (c) part of shown in figure 5;

Fig. 5: The person identification system.

V. RESULTS AND DISCUSSIONS

To evaluate the performance of the fingerprint verification algorithm developed, the MSE model in [18] and Regression was
used and the result of the MSE is presented in figure 6;

Fig. 6: MSE result of the fingerprint verification algorithm

IJISRT22AUG639 www.ijisrt.com 1047

Volume 7, Issue 8, August – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
MSE analysis aimed to measure the training error of showed that the error achieved is approximately zero, which
the system and determine how reliable it is when deployed is good. The regression result of the training process was
as a verification system. The result of the MSE shows an also used to evaluate the performance and presented in
error function of 0.056845Mu. The implication of this result figure 7;

Fig. 7: Regression of the fingerprint verification system

The ideal regression (R) value is 1, however when R is in [19] and the average R results achieved is 0.9891 and
approximately 1, then the system is said to be very good. MSE OF 0.050877Mu.
From figure 7, it was observed that the overall regression
result achieved using the mean of the multisets is 0.9999, Having tested the verification algorithm and validated
which is approximately 1. The implication of the result the results, the system was integrated as an expert system
showed that the verification algorithm was able to identify and tested using the fingerprint of someselected staff of
individuals based on their fingerprints. The results were ESUT institution, and the results are presented in figure 8
validated using a tenfold cross-validation technique model and 9 respectively showing the query fingerprint data and
the verification result;

IJISRT22AUG639 www.ijisrt.com 1048

Volume 7, Issue 8, August – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Fig. 9: Query staff fingerprint data

Fig. 10: The fingerprint verification result

VI. CONCLUSION REFERENCES

The need for a technological means of human [1.] Akeem T., Momoh I., Danlami J. (2016)”
identification has been identified by the world health Assessment of the variation of ghost employee fraud
organization [9] as the only reliable means to curb fraud in in Nigerian; 2008-2015
our society. This research has developed an artificial [2.] Albrecht CO (2009). International fraud: a
intelligence system that can verify the identity of individuals management perspective. Ph.D. dissertation.
using fingerprint data to help address the problem of payroll Barcelona: Universitat Ramon Lull.
fraud in our tertiary institutions. The system was tested and [3.] Amoako-Tuffour J (2002). Ghost names, shadow
the result showed that it was able to verify a person's workers, and the public sector wage bill. Accra,
identity with approximate zero (0) MSE which is very good. Ghana: Institute of Economic Affairs.
The study, therefore, recommends that other institutions and [4.] Editorial, Vanguard newspaper” The ghost worker
related stakeholders should adopt this system and solve the syndrome in Nigeria” 2 January 2021.
problem of payroll fraud once and for all. In this work, an [5.] Penn E., Mbuh N. Nchise D., Alain V. (2020) “
intelligent fingerprint verification system was developed to Employee fraud: an empirical review of ghost
solve the problem of payroll fraud and ghost worker workers” IJRAS; Vol 7; Issue 3; ISSN 2394-4404
syndrome. [6.] Agboola T., (2018) “ Effectiveness of Integrated
Payroll and Personnel Information System (IPPSS) in

IJISRT22AUG639 www.ijisrt.com 1049

Volume 7, Issue 8, August – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Nigerian public sector; Global journal of human [16.] Asogwa T.C and Asogwa T.C. (2018) “The
social science Application of Machine Learning For Digital
[7.] Lindelöw M (2008). Tracking public money in the Recognition of Identical Twins To Support Global
health sector in Mozambique: conceptual and Crime Investigation”; International Journal of
practical challenges. In: Amin S, Das J, Goldstein M, Computer Science and Engineering (IJCSE), Vol4
editors. Are you being served? New tools for pp. 427–433
measuring service delivery. Washington (DC): [17.] Olatinwo,(2014), Development of a Networked
International Bank for Reconstruction and Thumb Print-Based Staff Attendance Management
Development/World Bank. System, Department of Computer Science and
[8.] Lou Y, Wang M (2009). Fraud risk factor of the Engineering, Ladoke Akintola University of
fraud triangle assessing the likelihood of fraudulent Technology, Ogbomoso, Nigeria, American Journal
financial reporting. Journal of Business & Economics of Engineering Research (AJER) e-ISSN: 2320-0847
Research. 7(2):61–78. p-ISSN: 2320-0936 Volume-03, Issue-03, pp-121-
[9.] World Health Organization (2020) “Findings from a 126.
rapid review of literature on ghost workers in the [18.] Ene P.C and Ebere U.C (2022) “Development of An
health sector: towards improving detection and Intelligent Technique For Fingerprint And Face
prevention; ISBN 978-92-4-000434-4; ISBN 978-92- Recognition System” Enugu State University of
4-000435-1 (print version) Science and Technology (ESUT); Research on
[10.] Nyaledzigbor G (2015). Payroll fraud: effects of Intelligent Image Processing.
ghost names on the government wage bill in Ghana. [19.] KULKARNI, S.: Fingerprint feature extraction and
Ph.D.dissertation.Minneapolis (MN): Walden classification by learning the characteristics of
University. fingerprint patterns, Neural Network World (2011),
[11.] Obara C, Nangih E, Agba JN (2017). Accounting 21 (2011), no. 3, 219–226.
systems and payroll fraud in the public sector: a [20.] BART˚UNˇEK, J. S., J. S.—NILSSON, M.—
survey of selected ministries and parastatals in Rivers NORDBERG, J.—CLAESSON, I.: Neural network-
State, Nigeria. Journal of Accounting and Financial based minutiae extraction from skeletonized
Management. 3(2):10–24.T fingerprints, in TENCON 2006, IEEE Region 10
[12.] Md. Shakil, Rabindra Nath Nandi (2013) “ Conference (2006), 4 p.
Attendance Management System For Industrial [21.] IGEL, CH.—H¨USKEN, M.: Improving the
Worker Using Fingerprint Scanner” Global Journal Rproplearning algorithm, in The Second
of Computer Science And Technology, Volume 13, International Symposium on Neural Computation
Issue 6 Version 1.0, Type: Double Blind Peer (NC 2000), ICSC Academic Press, 2000, pp. 115–
Reviewed International Research Journal, Publisher: 121.
Global Journals Inc.(USA), Online ISSN: 0975-4172 [22.] Renjie C., Freedman D., Karni Z., (2010) “Content-
% Print ISSN:0975-4350, aware image resizing by quadratic programming”;
[13.] Tolulope Awode, OluwagbemigaShoewu proceeding of 2010; IEEE CVPR Workshop on
Oluwabukola Mayowa Ishola, Segun O. Nordia Florida, 1-8
Olatinwo,(2014), Development of a Networked [23.] Lawrence O., Jefferey V. (1989) “An approach to
Thumb Print-Based Staff Attendance Management fingerprint filter design” pattern recognition; vol 22;
System, Department of Computer Science and Issue 1; 1989; pp 29-38
Engineering, Ladoke Akintola University of [24.] Palme L., Al-Tarawneh S., Dlay S., Woo W., (2008)
Technology, Ogbomoso, Nigeria, American Journal “ Efficient fingerprint feature extraction: Algorithm
of Engineering Research (AJER) e-ISSN: 2320-0847 and performance evaluation” CSNDSP08; 978-1-
p-ISSN: 2320-0936 Volume-03, Issue-03, pp-121- 4244-1876; PP 581-584
126. [25.] Eneh Princewill C., Eneh Innocent I.,
[14.] Adewole K. S, Abdulsalam S. O, Babatunde R. S. EgoigweSochima V., and Ebere Chidi U. (2019)”
Shittu T. M., and Oloyede M. O, (2014) Deep Artificial Neural Network Based Obstacle
“Development of Fingerprint Biometric Attendance Detection And Avoidance For A Non-Holonomic
System For Non- Academic Staff in a tertiary Mobile Robot” International Research Journal of
Institution,” (IISTE: E-Journals, Computer Applied Sciences, Engineering and Technology
Engineering and Intelligent Systems, ISSN: 2222- Vol.5, No.1; January-2019; ISSN (1573-1405); p –
1719(paper), ISSN: 2222-2863(online), Volume 5, ISSN 0920-5691
No 2, [26.] Nair Vinod and Hinton Geoffrey E. (2010). Rectified
[15.] Haider Mehraj, and Ajaz Hussain Mir (2020) “A linear units improve restricted Boltzmann machines.
Survey of Biometric Recognition Using Deep Proceedings of the 27th International Conference on
Learning” Department and Communication Machine Learning (ICML10). 2010:807–814.
Engineering, National Institute of Technology,
Srinagar, J&K, India-19006,

IJISRT22AUG639 www.ijisrt.com 1050

Volume 7, Issue 8, August – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Engr. Dr. Ene Princewill Chigozie is a researcher under the Instrumentation and Robotic Control group at the Exxon Mobile
Laboratory and a Lecturer in Electrical and Electronic Engineering Department, Faculty of Engineering, Enugu State University
of Science and Technology (ESUT). He obtained a B.Sc. Degree in Electrical and Electronic Engineering, M.Eng. and Ph.D.
Degree in Instrumentation and Control Engineering from EEE Dept. ESUT in 2006, 2012, and 2019 respectively. He is a member
of the Nigerian Society of Engineers (NSE), the Council for the Regulation of Engineering in Nigeria (COREN), and the Institute
of Electrical and Electronic Engineers (IEEE). His scientific interests include Artificial Intelligence in Robotic Control,
Mechatronics, Signal Processing, Computer Vision, and Power System Security. He has collaborated actively with researchers in
other disciplines of engineering.

IJISRT22AUG639 www.ijisrt.com 1051