ABSTRACT
This research presents the development of a Discrete Firefly Algorithm (DFA) based feature
selection scheme for improved face recognition. Discrete Cosine Transform (DCT) and Haar
wavelet based Discrete Wavelet Transform (DWT) were used for feature extraction, and
Nearest Neighbour Classifier (NNC) was used as classifier. Extracted features are mostly
discrete in nature and most of the optimization techniques used in feature selection are
continuous so DFA is employed for feature selection. The developed DFA based feature
selection scheme was tested on Olivetti Research Labs (ORL) and Yale databases, and was
compared with Firefly Algorithm (FA), Principal Component Analysis (PCA) and Linear
Discriminant Analysis (LDA) respectively. The simulation was carried out in MATLAB
R2013b simulation environment, and the result obtained from ORL database for DFA
showed that the recognition accuracy (R.A) was found to be 97.75 % and recognition time
(R.T) was 42.27 seconds while for FA, the R.A was found to be 95.53% and R.T was 49.71
seconds. For the Yale database, the DFA had a R.A of 89.30% and a R.T of 40.33 seconds,
for FA, the R.A was 85.33% and R.T was 43.65 seconds. On applying DFA on local images
the R.A and R.T was 72.02% and 25.89 seconds respectively. The improvements in terms of
R.A and R.T of this system when comparing DFA with FA on ORL database were 2.27% and
14.97%, while the improvements on Yale database were 4.45% and 7.61% respectively. Also,
when compared with PCA, it gave an improvement of 25.48% in R.A and 23.82% in R.T,
while for LDA it gave an improvement of 38.84% in R.A and 27.81% in R.T for ORL
database. Also for the Yale database, when compared with PCA, it gave an improvement of
23.11% in R.A and 16.21% in R.T, and for LDA, it gave an improvement of 26.61% in R.A
and 20.01% in R.T respectively.
TABLE OF CONTENTS
DECLARATION I
CERTIFICATION II
DEDICATION III
ACKNOWLEDGEMENT IV
ABSTRACT VI
TABLE OF CONTENTS VII
LIST OF FIGURES XII
LIST OF TABLES XI
LIST OF APPENDICES
XERROR! BOOKMARK NOT DEFINED.
LIST OF ABBREVIATIONS 11
CHAPTER ONE
INTRODUCTION
12
1.1 Background of Research 12
1.2 Motivation 3
1.3 Significant of Research 4
3
1.4 Statement of Problem 4
1.5 Aim and Objectives 5
1.6 Methodology 5
1.7 Dissertation Organisation 7
CHAPTER TWO
LITERATURE REVIEW 9
2.1 Introduction 9
2.2 Review of Fundamental Concepts 9
2.2.1 Face Recognition 9
2.2.1.1 Face Recognition Design 11
2.2.2 Feature Extraction 13
2.2.2.1 Local binary pattern (LBP)
13
2.2.2.2 Gabor Filter
14
2.2.2.3 Histogram of oriented gradients (HOG)
14
2.2.2.4 Discrete cosine Transform (DCT) 15
2.2.2.5 Discrete Wavelet Transform (DWT) 16
2.2.3 Feature Selection. 19
2.2.3.1 Genetic Algorithm (GA) 21
2.2.3.2 Particle Swarm Optimization (PSO) 21
2.2.3.3 Ant Colony Optimization (ACO) 22
4
2.2.3.4 Cuckoo Search (CS) 22
2.2.3.5 Firefly algorithm (FA) 22
2.2.3.6 Discrete Firefly algorithm (DFA) 25
2.2.4 Classification or recognition 27
2.2.4. 1 Support Vector Machine (SVM). 28
2.2.4.2 Hidden Markov Model (HMM) 28
2.2.4.3 Back propagation neural network (BPNN).
29
2.2.4.4 Self Organizing Map (SOM).
29
2.2.4.5 Nearest Neighbor Classifier (NNC)
29
2.2.4.5.1 Recognition accuracy 30
2.3 Review of Similar Works 32
CHAPTER THREE
MATERIALS AND METHODS 41
3.1 Introduction 41
3.2 Implementation of Firefly Algorithm for Feature Selection for Face Recognition 41
3.2.1 Image acquisition 41
3.2.2 ORL face database 42
3.2.3 Yale face database 42
3.2.4 Local Images Face Database. 43
5
3.2.5 Importing of the Images into MATLAB („imread‟) 44
3.2.6 Converting RGB into Grayscale Images 44
3.3 Extraction using DCT and DWT 45
3.4 Feature Selection with FA. 46
3.4.1 Initialization of the FA Parameters 46
3.4.2 Initialization of Light Intensity. 46
3.4.3 Generating New Solution by Updating the Position of the Firefly. 47
3.5 Feature selection with DFA 47
3.5.1 Generating new solution by updating the position of the firefly 48
3.6 Feature selection with LDA 49
3.7 Feature Selection with PCA 50
3.8 Classification using Nearest Neighbour Classifier (NNC) 51
3.9 Performance Evaluation. 52
3.9.1 Recognition Accuracy 52
3.9.2 Recognition Time 52
3.9.3 Percentage Improvement 53
CHAPTER FOUR
RESULTS AND DISCUSSIONS 54
4.1 Introduction 54
4.2 Result of Gray Scale Conversion of Images on ORL, Yale and Local Images 54
6
4.3 Result of FA on ORL Database 56
4.4 Result of DFA on ORL Database 56
4.5 Result of FA on Yale Database 57
4.6 Result of DFA on Yale Database 57
4.7 Comparison of Results 58
4.7.1 Comparison of the Recognition Accuracy for DFA, FA, PCA and LDA on ORL
Database 58
4.7.2 Comparison of the Recognition Time of DFA, FA, PCA and LDA on ORL
Database 59
4.7.3 Comparison of the Recognition Accuracy for DFA, FA, PCA and LDA on Yale
Database 59
4.7.4 Comparison of the Recognition Time of DFA, FA, PCA and LDA on Yale
Database 60
4.8 Result of DFA on local Images 61
4.9 Performance Evaluation 62
4.9.1 Percentage Improvement on ORL Database 62
4.9.2 Percentage Improvement on Yale Database 62
CHAPTER FIVE
CONCLUSION AND RECOMMENDATION 64
5.1 Summary 64
5.2 Conclusion 64
7
5.3 Significant Contributions 64
5.4 limitation 65
5.5 Recommendations for Further Work 65
REFERENCES 66
CHAPTER ONE
INTRODUCTION
1.1 Background of Research
In the field of pattern recognition, face recognition has attracted interest from researchers due
to its numerous applications (public security, law enforcement and commerce, credit card
verification, criminal identification, access control, human-computer intelligent interaction,
digital libraries and information security) (Bakshi & Singhal, 2014). Face recognition is a
process of identifying or verifying a person‟s identity by matching input face biometrics as
against pre-defined faces in a database (Zhou et al., 2014).
In day to day social activities and interactions, the face seems to be an important factor for
easy identification (Shivdas., 2014.). Face recognition has advantages over the traditional
methods of identification, which involved the use of passwords and personal identification
number that provides accuracy and its case sensitiveness (Angle et al., 2005; Kaur & Singh,
2015), and it also offers non-contact process, captured or videoed easily, and provides
reliable face matching, and offers a wide range of applications (Bakshi & Singhal, 2014).
The face acts as a key factor of consideration in the public domain, playing a foremost
function in conveying uniqueness and emotion (Maini & Aggarwal, 2009). Face recognition
works basically in three stages which comprises of detection, feature extraction and
classification or recognition (Maini & Aggarwal, 2009). And the choice of approach to each
2
of these stages is vital to attaining better recognition accuracy. The face recognition problem
is complicated by age, skin, colour, gender, differing image qualities, facial expressions,
background, and illumination condition (Bakshi & Singhal, 2014).
In face detection, the goal is to find an object in an image as a face candidate that its shape
resembles the shape of a face (Saleh, 2009). In other words, it can be regarded as a process of
automatically detecting a face from a complex background to which the face recognition
algorithm can be applied. Researchers use pre-processing at this stage (Agarwal & Bhanot,
2015).
In feature extraction, high level information about individual patterns (like eyes, nose, lips,
etc.,) to facilitate recognition is extracted (Hemalatha & Govindan, 2015 ). Selection of
feature extraction method is probably the single most important factor in achieving high
recognition performance (Saleh, 2009). Approaches used for face extraction include discrete
cosine transform (DCT) (Hemalatha Gayatri & Govindan, 2015; Jadon et al., 2015), gabor
filter (Keche et al., 2015; Ruan et al, 2010), principal component analysis (PCA) (Bakshi &
Singhal, 2014; Satone & Kharate, 2014; Sawalha & Doush, 2012), local binary pattern (LBP)
(Babatunde et al., 2015), and discrete wavelet transform (DWT) (Kallianpur et al., 2016;
Manikantan et al., 2012).
In the classification or recognition stage, face samples are compared or matched with existing
known faces in the database (Richa & Josan, 2013). Some methods reported at this stage are
support vector machine (SVM) (Satone & Kharate, 2014; Xu & Lee, 2014), Hidden Markov
Model (HMM) (Jameel, 2015), Nearest Neighbour Classifier (NNC) (Agarwal & Bhanot,
2015) Back Propagation Neural Network (BPNN) (Shivdas., 2014.) and self-organizing map
(SOM) (Bakshi & Singhal, 2014).
The feature selection process involves determination of a feature subset which can best
represent a given feature set (Manikantan et al., 2012). Feature selection problem, is
3
challenging due to its combinatorial nature. Feature selection phase of the face recognition
process attempts to obtain the most discriminative features between two or more individual’s
faces to produce the best accuracy in databases capturing variations in illumination, pose,
expression or occlusion (Agarwal & Bhanot, 2015). Many among the large features are not
vital features and as such it causes over fitting of the face data and consequently reduce
performance of the system (Agarwal & Bhanot, 2015). Some of the optimization technique
used in feature selection are particle swarm optimization (PSO) (Hemalatha & Govindan,
2015; Ramadan & Abdel-Kader, 2009; Unler & Murat, 2010; Xue et al., 2014), firefly
algorithm (Agarwal & Bhanot, 2015), genetic algorithm (GA) (Boubenna & Lee, 2016;
Satone & Kharate, 2014), ant colony optimization (ACO) (Babatunde et al, 2015; Kanan &
Faez, 2008) and artificial bee colony (ABC) (Kallianpur et al., 2016; Khan & Gupta, 2016).
However, these algorithms are continuous and requires continuous problem thus the face
recognition which is discrete requires to be converted to continuous or the algorithm is
converted to discrete which is time consuming hence this research hopes to employ a discrete
firefly algorithm (DFA) for feature selection to address this challenge in order to improve
recognition accuracy.
1.2 Motivation
Face recognition is a process of identifying or verifying a person‟s identity by matching input
face biometrics as against pre-defined faces in a database. The steps in face recognition are
feature extraction, feature selection and classification. Feature selection process involve
determination of a feature subset which can best represent a given feature set. Previous,
researchers have proposed many metaheuristic search algorithm and hybrid for feature
selection, which are continuous as such require more Recognition time and reduces
recognition accuracy, thus feature selection which consist of extracted feature which are
4
discrete require discrete algorithms which lead to an efficient Recognition time and
recognition accuracy. Thus, this research work offers the use of a metaheuristic algorithm
known as discrete firefly algorithm (DFA) for feature selection to address this problem.
1.3 Significance of Research
The significance of this research is the development of a discrete firefly algorithm based
feature selection scheme for improved face recognition, which has added capabilities to the
standard feature selectors in face recognition through the minimization of the number of
features that improve performance prediction of recognition rate. This has not been
considered by the previous researchers.
1.4 Statement of Problem
Face recognition is a process of identifying or verifying a person‟s identity by matching input
face biometrics as against pre-defined faces in database. It finds application in areas such as
public security, law enforcement, credit card verification, criminal identification, access
control, human-computer intelligent interaction and information security. The steps in face
recognition are detection, feature extraction, feature selection and classification. Feature
selection problem, is challenging due to its combinatorial nature. Feature selection phase of
the face recognition process attempts to obtain the most discriminative features between two
or more individual’s faces to produce the best accuracy in databases capturing variations in
illumination, pose, expression or occlusion. However, the problem of over fitting of the face
data results in reduced performance of the system. Some of the optimization technique used
in feature selection are PSO, FA, GA, ACO and CSA. However, these algorithms are
continuous and requires continuous problem thus the face recognition which is discrete
requires to be converted to continuous or the algorithm is converted to discrete which is time
5
consuming hence this research hopes to employ a Discrete Firefly Algorithm to address this
challenge.
1.5 Aim and Objectives
The aim of this research work is to develop an improved face recognition based on DFA for
feature selection. To achieve this aim, the objectives are as follows:
1. To implement and evaluate the performance of the face recognition system (FRS)
using firefly algorithm (FA) and discrete firefly algorithm (DFA) respectively for
feature selection in terms of Recognition time and recognition accuracy using the
Olivetti Research Labs (ORL) and Yale facial images databases.
2. Comparison of the performance of DFA based system with, linear discriminant
analysis (LDA) and principal component analysis (PCA) based system on (1).
3. To apply the DFA based system on faces of 10 people with 5 different facial
expressions collected at Zaria city of Kaduna state.
1.6 Methodology
The methodology to be adopted for this research are as follows:
1. Obtaining images that are used in the research as follows:
a. Benchmark face database of Olivetti Research Labs (ORL) and Yale from
http://www.uk.research.att.com/facedatabase.html and
http://cvc.yale.edu/projects/yalefaces/yalefaces.html respectively.
b. Local images in Zaria city of Kaduna state to create a database for application
using the developed method.
2. Implementation of the face recognition system based on Firefly Algorithm (FA) for
feature selection in MATLAB 2013b as follows:
6
a. Importing of the images into MATLAB 2013b simulation environment and
converting them into grayscale images.
b. Extraction of all the images using Discrete Cosine Transform (DCT) and Discrete
Wavelet Transform (DWT).
c. Carrying out feature selection with FA as follows:
i. Initializing the FA parameters population X and light absorption
coefficient ( ) .
ii. Initializing the light intensity i (I ) .
iii. Generating new solution by updating the position of the firefly j i (I I ) .
iv. Evaluating new solution and updating the light intensity.
d. Classification using Nearest Neighbour Classifier (NNC).
3. Implementing the face recognition system using Discrete Firefly Algorithm (DFA) for
feature selection and comparing the performance in (2) on the benchmark face
database of Olivetti Research Labs (ORL) and Yale.
a. Pre-processing and importing of the images into MATLAB 2013b simulation
environment and converting them into grayscale images.
b. Extraction of all the images using Discrete Cosine Transform (DCT) and Discrete
Wavelet Transform (DWT).
c. Carrying out feature selection with DFA as follows:
i. Initializing the DFA parameters population X and light absorption
coefficient ( ) .
ii. Initializing the light intensity i (I ) .
iii. Generating new solution by updating the position of the firefly j i (I I )
using hamming distance, lengths of the firefly and movement function.
7
iv. Evaluating new solution and updating the light intensity.
d. Classification using Nearest Neighbour Classifier (NNC).
4. Carrying out feature selection with LDA as follows:
i. Create a D-dimensional samples (1 (2 (N X ,X ,…,X
ii. Find a measure of separation of the two classes, within class scatter and
between class scatter.
iii. Find the optimum vector W.
5. Carrying out feature selection with PCA as follows:
i. Create A matrix from training images and Compute B matrix from A.
ii. Compute eigenvectors of C from eigenvectors of B.
iii. Select few most significant eigenvectors of C for face recognition.
iv. Compute coefficient vectors corresponding to each training images.
v. For each person, coefficient will form a cluster, compute mean of mean
cluster.
6. Comparison of the DFA, FA, LDA and PCA in terms of Recognition Time and
Recognition Accuracy
7. Application of the DFA based approach on the faces of 10 people with 5 different
facial expressions each, collected at Zaria city of Kaduna state as implemented in (3).
1.7 Dissertation Organization
The general introduction has been presented in Chapter One. The rest of the chapters are
structured as follows: First, detailed review of related literature and relevant fundamental
concepts about face recognition, discrete cosine transform (DCT), discrete wavelet
transform (DWT), firefly algorithm (FA), discrete firefly algorithm (DFA) for feature
selection and Nearest Neighbour Classifier (NNC) are carried out in chapter two. Second,
8
an in depth approach and relevant mathematical models describing the development of an
improved face recognition base on discrete fire fly for feature selection. Third the
analysis, performance and discussion of the result are shown in chapter four. Finally,
conclusion and recommendations of further work makes up the chapter five. Finally list
of cited references and MATLAB codes in the appendices are provided at the end of this
dissertation.
Do you need help? Talk to us right now: (+234) 08060082010, 08107932631 (Call/WhatsApp). Email: [email protected].
IF YOU CAN'T FIND YOUR TOPIC, CLICK HERE TO HIRE A WRITER»
