FINGER-VEIN RECOGNITION SYSTEM

(1)

FINGER-VEIN RECOGNITION SYSTEM

BY

SUBHA FAIRUZ

A dissertation submitted in fulfillment of the requirement for the degree of Master of Science (Computer and Information

Engineering)

Kulliyyah of Engineering

International Islamic University Malaysia

APRIL 2019

(2)

ii

ABSTRACT

Biometric feature-based recognition systems are trending on the data security business.

The unique features of the human body are used in security systems also known as a biometric system, such as voice, fingerprint, iris recognition, face recognition, etc.

Among all, finger-vein recognition is at the peak of its popularity amidst the consumers, as well as security researchers and organizations are willing to use this technology commercially. However, finger-vein recognition is exclusive because of the veins which are underneath of human’s skin rather than outside structure. Thus, compared to other biometric technology finger-vein is different plus it is almost impossible to fabricate, even two twins’ finger-veins are fully different from each other. To establish a finger- vein recognition system, it required a huge amount of finger-vein data and for that reason, a finger-vein acquisition device is made as well as created the dataset. To create the finger-vein acquisition device, near infrared (NIR) imaging technology had been used and built a dataset which is split into two datasets, one for training and another for testing. To train the dataset convolutional neural network (CNN) is used via transfer learning of Alexnet using MATLAB in order to finding out the recognition accuracy.

After that, a real-time finger-vein recognition system is developed which captured finger-vein images of individuals and after running those images through the recognition system generates a recognition rate. This research obtained 100% accuracy when tested with the stored dataset, however, in real-time 99%, predictive accuracy is received in several experiments which is notably a satisfactory result.

(3)

iii

ثحبلا ةصلاخ

رعتلا ةمظنأ ترشتنا صئاصلخا ىلع ةدمتعلما )ةيويلحا( ةيترمويبلا ف

ةينامسلجا (Biometric feature-based recognition

systems) ةمظنلأبا اًضيأ فرعت تيلاو ةينملأا ةمظنلأا في ةديرفلا ةينامسلجا صئاصلخا مدختست .تناايبلا ةملاس نمضتت تيلا لامعلأا في

.مهيرغو هجولا قيرط نعو ،ينعلا ؤبؤب قيرط نع فرعتلاو ،عبصلإا ةمصبو ،توصلاك ةيترمويبلا ةمدختس

ُلما ةيترمويبلا صئاصلخا عيجم نمض نم

سؤلما حمطتو ،نملآا لامج يثحباو ينكلهتسلما ينب اًراشتنا رثكلأا وه عبصلإا قرِع ةمصب مادختسبا فُّرعتلا نإف ،قيثوتلا وأ ناملأا في تاس

لإ يرصح وه عبصلإا قرِع قيرط نع فرعتلا نإف ،كلذ عمو .يراتج لكشب ةينقتلا هذه مادختسا لىإ سيلو ناسنلإا دلج تتح عقي قرِعلا ن

،ىرخلآا ةيترمويبلا صئاصلخبا ةينقتلا هذه ةنراقم دنع ،كلذلو ،جرالخا في سانلا عباصأ قورع نإف كلذ ىلع ةولاع ،فلتمخ عبصلإا قرِع نإف

.ضعبلا اهضعب نع فلتتخ مئاوتلا عباصأ قورع تىح هنلأ كلذو ،اهريوزت ليحتسي ع دمتعم نيمأ ماظن ءانبل

،عبصلإا قرِع قيرط نع فُّرعتلا ىل

باستكا زاهج عنصلو .هتناايبو عبصلإا قرع باستكا زاهج عنص تم ،ببسلا اذلهو ،عباصلأا قورع نم ةمخض تناايب ةعوملمج جايتحا كانهف ءارملحا تتح ةبيرقلا ةعشلأبا يفيطلا ريوصتلا ةينقت مادختسا تم عبصلإا ضرع (NIR)

هميسقت تم جذونم ءانب في لىإ

،تناايبلل ينتدعاق

ةيفيفلاتلا ةيبصعلا ةكبشلا مادختسا تم ،تناايبلا ةعوممج بيرجتلو .رابتخلال ىرخلآاو بيردتلل ةدحاو (

convolutional neural

network ب صالخا ملعتلا لقن قيرط نع ،)

AlexNet مادختسبا

MATLAB ،كلذ دعب .فُّرعتلا ةقِد ىلع لوصحلل

ريوطت تم

يقح نيمز ماظن اذه .فُّرعتلا ةبسن ديلوت تم فُّرعتلا ماظن ىلع اهليغشت دعبو صاخشلأا قورعل اًروص طقتلي ناك يذلاو قرِعلا ىلع فرعتلل يق

ةقد ةبسن ىلع لصح ثحبلا 100

ةيؤبنت ةقِد ةبسن ىلع يقيقلحا نمزلا في انلصح دقف ،كلذ مغر ،ةنزخ

ُلما تناايبلا ةدعاق في هرابتخا دنع %

انتبر ُتج في99%

جئاتن يهو .ةيضرُم

(4)

iv

APPROVAL PAGE

I certify that I have supervised and read this study and that in my opinion, it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Master of Science (Computer and Information Engineering).

………..

Mohamed Hadi Habaebi Supervisor

………..

Farhat Anwar Co-Supervisor

I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Master of Science (Computer and Information Engineering).

………..

Md Rafiqul Islam Internal Examiner

………..

S.M.A. Motakabber Internal Examiner

This dissertation was submitted to the Department of Electrical and Computer Engineering and is accepted as a fulfilment of the requirement for the degree of Master of Science (Computer and Information Engineering).

………..

Mohamed Hadi Habaebi Head, Department of Electrical and Computer Engineering

This dissertation was submitted to the Kulliyyah of Engineering and is accepted as a fulfilment of the requirement for the degree of Master of Science (Computer Information Engineering).

………..

Ahmad Faris Ismail

Dean, Kulliyyah of Engineering

(5)

v

DECLARATION

I hereby declare that this dissertation is the result of my own investigations, except where otherwise stated. I also declare that it has not been previously or concurrently submitted as a whole for any other degrees at IIUM or other institutions.

Subha Fairuz

Signature ... Date ...

(6)

vi

INTERNATIONAL ISLAMIC UNIVERSITY MALAYSIA

DECLARATION OF COPYRIGHT AND AFFIRMATION OF FAIR USE OF UNPUBLISHED RESEARCH

FINGER-VEIN RECOGNITION SYSTEM

I declare that the copyright holders of this dissertation are jointly owned by the student and IIUM.

No part of this unpublished research may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise without prior written permission of the copyright holder except as provided below

1. Any material contained in or derived from this unpublished research may be used by others in their writing with due acknowledgement.

2. IIUM or its library will have the right to make and transmit copies (print or electronic) for institutional and academic purposes.

3. The IIUM library will have the right to make, store in a retrieved system and supply copies of this unpublished research if requested by other universities and research libraries.

By signing this form, I acknowledged that I have read and understand the IIUM Intellectual Property Right and Commercialization policy

Affirmed by Subha Fairuz

……..……….. ………..

Signature Date

(7)

vii

ACKNOWLEDGEMENTS

Firstly, it is my complete pleasure to dedicate this work to my dear parents (A. A. M.

Wares and Rokshana Pervin) and my siblings, acquaintance, who allowed me the gift of their constant trust in my ability to fulfil this goal: thank you for your support and patience.

I wish to state my admiration and thanks to those who provided data for my work.

I also would like to thank all the members of my dissertation committee, thank you for sticking with me.

Finally, a special thanks to my beloved supervisor Associate Prof. Dr. Mohamed Hadi Habaebi for his continuous support, encouragement, and leadership regarding this work, for which I will be forever grateful.

(8)

viii

LIST OF TABLES

Table 2.1 Related Work Comparison Table 24

Table 4.1 Confusion Matrix 41

Table 4.1 Comparison between real-time finger-vein images 24 Table 4.2 Training Parameter and Result of Experiment 1 44 Table 4.3 Training Parameter and Result of Experiment 2 47 Table 4.4 Training Parameter and Result of Experiment 3 49 Table 4.5 Training Parameter and Result of Experiment 4 50 Table 4.6 Comparison between Direct-data and Pre-processed-data 53 Table 4.7 Comparison of Recognition Rate between pieces of

research works and this work

54

(11)

xi

LIST OF FIGURES

Figure 1.1 Layers of human skin 2

Figure 1.2 Spreading light at different wavelength inside tissue 2

Figure 1.3 Overview of the Methodology 6

Figure 2.1 The recognition rate with different numbers of the images for training

12

Figure 2.2 ROC curves for the three methods 12

Figure 2.3 Recognition rate curves in multi-sample 13

Figure 2.4 The FAR and FRR Curves of the Method Combining the wavelet Transformation and Energy Feature

15

Figure 2.5 Recognition rate curves in multi-sample

16

Figure 2.6 Recognition rate curves in multi-sample 16

Figure 2.7 Architecture of AlexNet 17

Figure 2.8 The proposed CNN design Integrate 18

Figure 2.9 Comparison between CNN and other algorithms in SDUMLA-FV database

20

Figure 2.10 ROC curve of different method 20

Figure 2.11 CNN architecture design 22

Figure 2.12 ROC curves of finger-vein recognition in low- quality databases 23 Figure 3.1 Finger-vein Identification System’s Flowchart 28 Figure 3.2 Visual Work-flow of Finger-vein Recognition 29 Figure 3.3 An example of image acquisition of finger-vein by this device 29 Figure 3.4 Create a database of Finger-vein images by capturing images from

the device

30 Figure 3.5 Finger-vein images recognition System Method 31

Figure 3.6 Different Steps of Pre-processing 32

Figure 3.7 Pictographic representation of max-pooling layer 35 Figure 3.8 Transfer learning of Alexnet of finger-vein recognition system 37 Figure 3.9 Predicting 100% Right Hand Index Finger of P1 38 Figure 3.10 Predicting 92% Right Hand Middle Finger of P1 39

(12)

xii

Figure 4.1 Best Case of Separability (AUC=1) 44

Figure 4.2 Worst Case of Separability (AUC=0) 44

Figure 4.3 Case of Inability to Separate [Type-1] (AUC=0.5) 45 Figure 4.4 Case of Inability to Separate [Type-2] (AUC=0.7) 45

Figure 4.5 ROC curve of Experiment 1 47

Figure 4.6 Training Accuracy and Loss of Experiment 1 47

Figure 4.7 Testing Real-time finger-vein images with trained data 49

Figure 4.10 False test data 51

Figure 4.14 Testing Real-time finger-vein images with pre-processed 54 trained data

Figure 4.15 Right Hand Middle Finger of P1 Shows 98% Recognition Rate in Real-time

57 Figure 4.16 Right Hand Index Finger of P1 Shows 100% Recognition Rate in

Real-time

57

(13)

xiii

LIST OF ABBREVIATIONS

2D Two Dimensional

AUC Area Under Curve

CCD Charged Coupled Device

CCTV Closed-Circuit Television

CLAHE Contrast Limited Adaptive Histogram Equalization CMOS Complementary Metal-Oxide Semiconductor

CNN convolutional neural network

CPU Central Processing Unit

FPR False Positive Rate

GPS Global Positioning System

GPU Graphics Processing Unit

IR Infrared

LED Light emitting diode

mm Millimeter

NIR Near Infrared

nm Nano Meter

ONPP Orthogonal Neighborhood Preserving Projections

PCA Principal Component Analysis

ReLu Rectified Linear Unit

ROI Region of Interest

ROC Receiver operating characteristic

TPR True Positive Rate

USB Universal Serial Bus

(14)

xiv

LIST OF SYMBOLS

π Pi - the ratio circumference and diameter of a circle σ Sigma-The standard deviation of the distribution Σ Summation- sum of all the values in range of series

ΣΣ Double Summation

V Transform matrix

W Weight

b Bias

O Output size

F(x) Maps values of x to f(x)

(15)

1

CHAPTER ONE INTRODUCTION

1.1 BACKGROUND

The research area of private information security has become the center of attention due to the exceptional development of recognition systems dependent on human biometric attributes which are distinctive, for example, finger-vein, retina, face and so forth.

Finger vein recognition is more secure than other authentication systems since finger- veins of any two people (even twins) are not same, over time grownup people’s finger vein remains same as well as it varies from person to person's fingers (F. Tagkalakis and V. Fotopoulos, 2015). Besides, finger-vein is an internal biometric feature in humans and due to this fact, the odds of information theft or loss is near zero. Among all the biometric traits of a human, finger-vein is proven to be the most promising. The authentication system works by contrasting vein patterns against pre-stored patterns of subjects in the database. The network of veins underneath the finger skin come into visibility due to the movement of the blood vessel (A. Agarwal, S. Maheshwari, and G.

Yadav, 2014).

When it comes to acquiring an individual’s vein image from a finger, lighting conditions play a key role and this finger vein image acquisition technique is non-intrusive and it is carried out using Near-Infrared light (NIR). The underlying reason for using NIR imaging is the low visibility of finger veins in normal lighting conditions which makes the image acquisition process difficult. Human skin consists of three types of layers known as epidermis, dermis, and hypodermis; Figure 1.1 shows the skin part. The

(16)

2

epidermis contains no blood vessels though it is the first point of contact with the imaging device. Diffusion occurs in the second layer, which is essential for the supply of nutrition and hypodermis consists of arteries, nerves, veins, and fat cells. Depths of hypodermis and dermis ranges from 0.6 to 3 mm.

Figure 1.1: Layers of human skin (H. K.AlGhozali, Setiawardhana, and R. Sigit, 2016)

Figure 1.2: Spreading light at different wavelength inside tissue (Ravi Varma N., Sandip D. Sahane, 2014)

(17)

3

The highest thickness of the hypodermal layer can be 0-3 millimeter. Perception is essential for understanding the lighting arrangement which will reach into the considerable depths of a person's finger-vein (H. K. Al Ghozali, Setiawardhana, and R.

Sigit, 2016). The wavelength of the visible light ranges between 400 nanometers to 700 nanometers which travel through the tissue reaching distinct depths and spreads various wavelengths. Visible lights wavelength manages to reach the epidermal and dermal layers of the skin which is not convenient as veins exist in hypodermal layers. Blood vessels of this hypodermal layers contain hemoglobin which runs through the finger- veins. However, near-infrared light ranges between 760 nanometers to 850 nanometers hence it is able to reach the depths of the hypodermal layer where finger-veins exist.

These finger-veins can be captured by using a camera.

1.2 THE PROBLEM STATEMENT

Vein based biometric authentication system is both safe and secure to implement.

Finger-veins are an internal bodily feature of humans which is unique to every individual besides it is nearly impossible to fraud or mimic vein patterns. Although it is highly secure in physical terms when it comes to recognition, if the pre-processing of finger-vein images is not collected using the right method then it could generate erroneous results. It reduces the efficiency and accuracy of working with finger-vein recognition system. To ensure utmost efficiency and accuracy training based convolutional neural network can be used.

A finger-vein recognition system includes acquiring finger-vein images which would later need processing to be able to use in recognition technique. Conventional non-train- based approaches use line tracking method on input images of finger-vein to extract the

(18)

4

features where there are chances of inaccurate detection which can lower the rate of finger vein recognition performance. Therefore, training-based learning method is proven to be much more efficient as a large volume of data can be used accurately in a better approach.

1.3 RESEARCH OBJECTIVES

The prime objective of this research work is to develop an efficient finger vein recognition system utilizing deep learning approaches. This can be achieved by the following precise objectives:

1. To study and analyses, the ongoing trend in finger-vein recognition approaches.

2. To design and implement a new finger-vein recognition system using the convolutional neural network.

3. To evaluate the developed system for ensuring proof of concept.

1.4 RESEARCH SIGNIFICANCE

The proposed system will immensely contribute to biometric security-based fields which are fine-tuned day by day in order to provide better security. Technology is implemented into every walk of life as days are passing by and distinct biometric features are used as a base for providing efficient and secure systems where finger-vein recognition is no less. Institutions such as banks are the most vulnerable where automated finger-vein recognition systems can be used to ensure utmost security by stopping forgery or use of fraudulent data. Moreover, with this work researcher with a similar field will discover new and improved real-time finger-vein recognition system which strengthens the biometric recognition territory. Thus, such continual research would result in a breakthrough system to be found.

(19)

5 1.5 RESEARCH METHODOLOGY

At first, a finger-vein image acquisition device is designed and built. After that, a dataset is created by collecting finger-vein images from several individuals, then those images are pre-processed before it is used in the training. Next, the convolutional neural network (CNN) is used to train and made a finger-vein recognition system. Then, examine the results, if satisfied it will go to the next step otherwise will return to device setting step, because creating a proper dataset is mandatory to recognize a finger-vein recognition system. In real-time finger-vein recognition, finger-vein matching is predicted by using CNN and depending on the accuracy of the predicted score recognition efficiency is determined. Finally, through training and testing approach, the finger-vein recognition system is analyzed, evaluated and achieved higher accuracy.

(20)

6

Figure 1.3 Overview of the Methodology Start

Design a finger-vein acquisition device

Develop a finger-vein recognition system

Examine the result and Algorithm

YES

NO

Simulate Finger-vein Recognition System

Analyze Performance and Evaluate

Validate the System

End

(21)

7 1.6 RESEARCH SCOPE

An affordable all-purpose security solution based on finger-vein recognition is provided to ensure the security and privacy of people which is an alternative to the pricey ones available in the market, especially in developing and underdeveloped regions where security breaches have become common activities. Both near-infrared lighting technology and deep learning method such as convolutional neural network functioned as surpassing scope for this research work recognizing finger-vein of individuals accurately. Moreover, implementing such an advanced system in real-time with higher accuracy ensured.

1.7 CONTRIBUTIONS

The research contributions are given below:

1. The near-infrared imaging technology is used as a basis for capturing finger- vein images of individuals as it is low-cost and easy to make and developed a finger-vein image capturing device.

2. As normal cameras are unable to capture images of finger-vein to capture the finger-vein images the camera was adjusted so that it could get images of finger- veins.

3. A prototype device was designed and developed to make finger-vein images dataset as publicly available data-sets are hard to get.

4. The convolutional neural network which is applied to match with the previous

(22)

8

dataset so that it can ensure the efficiency of the predicted value of a people's finger-vein.

5. A real-time automated finger-vein recognition system is developed which provided up to the level of 100% accuracy.

1.8 DISSERTATION ORGANIZATION

This dissertation is arranged with five different chapters. The first chapter begins with an introduction, an overview of this research, research objectives and research methodology. In the Second chapter, the literature review of related work is emphasized and detailed information regarding this research's subject area. The third chapter illustrates the methodology of this work step by step and chapter four is about the results and analysis of the experiments, performance and real-time recognition. Finally, the last chapter is concluded with some recommendations and future scope of this study.

(23)

9

CHAPTER TWO LITERATURE REVIEW

2.1 INTRODUCTION

Finger veins are an internal bodily feature of humans, that is not visible to naked eyes normally which is why it requires a special camera and lighting condition to be able to see the veins. Particularly, Near Infrared spectrum range is suitable for viewing the veins and cameras with certain abilities can be used to capture the vein image. In this segment contextual basics of the equipment and applications used in this research work are discussed in a comprehensible manner.

2.2 INFRARED LIGHT

Light is a kind of radiation belonging to the electromagnetic spectrum. In particular what appears as visible light to human eye resides between Ultraviolet and Infrared spectrum which in terms of wavelength ranges between 400 to 700nm. The infrared spectrum is divided into three main groups known as Near Infrared (NIR), Mid Infrared (MIR) and Far Infrared (FIR). In particular, NIR spectrum possesses a wavelength of 760nm to 850nm which creates illumination under the skin and visualizes what’s underneath to the camera in order to capture it.

2.3 NIR REFLECTION ON HUMAN SKIN

The human skin consists of two layers known as Epidermis and Dermis. There is a layer of fat cells called subcutaneous layer under the dermis layer which comprises of components like melanin, blood and keratin. When NIR light is shone on the outer surface of the skin it goes directly into the epidermis layer where some light is

(24)

10

permeated and spread into the next layer. In the dermis layer most of the light gets scattered by fat cells while a part of the light gets absorbed (Habib Khirzin Al Ghozali, Setiawardhana, 2016). The blood veins containing deoxygenated blood vessels absorbs this light and appears darker which in result helps capture vein image using the camera.

There are both veins and arteries present underneath the skin but only veins appear in the camera and the reason behind that is veins contain more deoxy-hemoglobin concentration (53%) while arteries comprise of more oxy- hemoglobin content (90%- 95%). (Mayur Wadhwani, Abhinandan Deepak Sharma, 2015)

2.4 RELATED WORK

Finger vein recognition poses two major challenges and they are, quality of the finger vein image and limited texture information of the finger veins. In order to solve these challenges a framework was proposed in this work which implemented a unified method using manifold learning and point-manifold distance function for finger vein recognition. Using manifold learning efficiently involves dimensionality reduction. An image is considered to be a point in a high dimensional space where each pixel of the image is represented as one dimension. However, working with finger vein images needs dimensionality reduction. There are a few techniques available for doing this reduction among which Principal Component Analysis (PCA) and Orthogonal Neighborhood Preserving Projections (ONPP) are mention worthy. One of the most widely used geometric data modelling technique is PCA (Jolliffe, 1986) which finds the linear subspace that best represents the input data. It can be formulated as,

𝑓 = (𝑥^𝑇𝑏_{1 ,}𝑥^𝑇𝑏2,………..,𝑥^𝑇𝑏_{𝑑 ,}) = 𝐶₁, 𝐶₂, … … … , 𝐶_𝑑 (2.1) Where X is an input data set, b1, b1,…..,bd is a set of basis images and a finite subset of R^d, f is a function which projects a point X in R^Dto a set of coefficients in R^d and C The