FACE DETECTION AND RECOGNITION

Project report submitted in partial fulfilment of the requirement for

the degree of Bachelor of Technology
In
Computer Science and Engineering/Information Technology

By
Karandeep Singh (141412)

Under the supervision of

Ms. Ruchi Verma
To

Department of Computer Science & Engineering and Information

Technology
Jaypee University of Information Technology, Waknaghat,
Solan- 173234, Himachal Pradesh
 CANDIDATES’ DECLARATION

We hereby declare that the work presented in this report entitled “Face Detection and
Recognition” in partial fulfilment of the requirements for the award of the degree of
Bachelor of Technology in Computer Science and Engineering submitted in the
department of Computer Science and Engineering, Jaypee University of Information
Technology, Waknaghat, is an authentic record of our own work carried out over a period
from January 2018 to May 2018 under the supervision of Ms. Ruchi Verma, Assistant
Professor, Department of Computer Science and Engineering.

The matter embodied in the report has not been submitted for the award of any other
degree or diploma.

Karandeep Singh (141412)

This is to certify that the above statement made by the candidate is true to the best of my
knowledge.

Ms. Ruchi Verma

Assistant Professor
Computer Science & Engineering
Dated:

i
 ACKNOWLEDGEMENT

We are grateful and indebted to Ms. Ruchi Verma, Department of Computer Science for
her help and advice in the completion of this project report.

We also express our deep sense of gratitude and appreciation to our guide for his constant
supervision, inspiration and encouragement right from the beginning of this
project.

We also want to thank our parents and friends for their immense support and confidence
upon us. We deem it a pleasant duty to place on record our sincere and heartfelt gratitude
to our project guide for his long sightedness, wisdom and co-operation which helped us in
tackling crucial aspects of the project in a very logical and practical way.

Karandeep Singh (141412)

ii
 TABLE OF CONTENTS

Candidates’ Declaration ........................................................................................................ i

Acknowledgement ............................................................................................................... ii
Table of Contents ...............................................................................................................iii
List of Figures ................................................................................................................................................ iv
List of Tables ....................................................................................................................... v
Abstract ............................................................................................................................... vi
1 Introduction ........................................................................................................... 1
1.1 Introduction ........................................................................................................ 1
1.2 Problem Statement ............................................................................................. 2
1.3 Objectives .......................................................................................................... 3
1.4 Methodology ...................................................................................................... 4
1.5 Organization ....................................................................................................... 4
2. Literature Survey ................................................................................................. 5
2.1 Face detection approaches ................................................................................ 5
2.2 Face Recognition Approaches .......................................................................... 5
2.3 Review Models ................................................................................................. 8
3 System Development .......................................................................................... 13
3.1 Design .............................................................................................................. 13
3.1.1Assumptions .................................................................................................. 13
3.2 Experiments and results ................................................................................... 20
3.2.1 System Hardware ......................................................................................... 21
3.2.2 System Software ........................................................................................... 21
4 Performance Analysis ......................................................................................... 22
4.1 Methods Used and System Design .................................................................. 22
4.1.1 Method 1 ...................................................................................................... 23
4.1.2 Method 2 ...................................................................................................... 24
4.1.3 Method 3 ...................................................................................................... 27
5 Conclusions ......................................................................................................... 28
5.1 Reflection ......................................................................................................... 28
5.2 Project Conclusion ........................................................................................... 28
5.3 Future Scope .................................................................................................... 29
6 References ........................................................................................................... 30

iii
 LIST OF FIGURES

2.1 Configuration of generic face recognition .................................................................... 6

2.2 HMM for Face Recognition .......................................................................................... 9
2.3 Block Extraction from Image ...................................................................................... 10
2.4 Structure of RBF Neural Network .............................................................................. 11
3.1 Face Detection and Recognition flow ......................................................................... 14
3.2 Example of taken/white balance ................................................................................. 11
3.3 Results of Segmentation on Uncorrected Left) and Corrected Image (Right) ............ 15
3.4 Result of filtering operations on face candidate ......................................................... 16
3.5 Facial Feature Extractions for the author .................................................................... 16
3.6. Algorithm of Face Recognition ................................................................................. 17
3.7 Binary to decimal ........................................................................................................ 18
3.8 Histogram .................................................................................................................... 19
3.9 Haar features ............................................................................................................... 19
3.10 Classifiers .................................................................................................................. 20

iv
 LIST OF TABLES

2.1 Difference between two types of features ...................................................................... 8

v
 LIST OF ABBREVIATIONS

S No. Abbreviations Description

1 IDE Integrated Development Environment

2 HMM Hidden Markov Model
3 RGB Red Green Blue
4 YCbCr Luma component, blue difference, red
Difference

vi
 ABSTRACT

Face recognition is an important application of Image processing owing to its use in many
fields. The project presented here will be developed after studies of various face
recognition methods and their efficiencies. An effective and real time face recognition
system based on OpenCV and C++ (and/or MATLAB) will be developed in this project
(Many other techniques are available to apply face recognition one of them is Principle
Component Analysis (PCA)).

This report describes our fourth year project “Face Detection and Recognition “. It
discusses the various approaches and techniques that have been used for detection and
recognition of faces. Furthermore, it mentions the various methods used for development
and their detailed description. It showcases the results obtained and the testing performed
to test the developed software artefact. This report also mentions the various tools and
APIs that can be used to implement various methods and techniques in these fields.

vii
 CHAPTER -1

INTRODUCTION

1.1 INTRODUCTION

Pictures expect an imperative part in the present information in light of the way that a
lone picture addresses a thousand words. “Face Detection” is said to be recognition of the
input inquiry as a face and figure out information in the data picture. “Face Recognition”
is basically whether the found face exists in the data as in whether or not it matches with
the face of a person or not.”Face Recognition” system ascends from the minute that
machine began to end up increasingly "keen" and had the progress of fill in, right or help
the absence of human capacities and faculties.

Face affirmation is one of just a modest bunch few biometric procedures that have the
advantages of both high exactness and low intruding. It has the exactness of a
physiological approach without being meddling. Over late years, various experts have
proposed differing face affirmation procedures, impelled by the extended number of bona
fide applications that require the acknowledgment of human countenances. There are a
couple of issues that make modified stand up to affirmation a to a great degree
troublesome task. The centrality of customized stand up to affirmation is much to adjust
to different assortments of photos of a comparable face in light of changes in the going
with parameters, for instance, act, light, appearance, movement, facial hair, glasses, and
foundation. Face affirmation advancement is well impel that can interface for some
business applications, for instance, individual recognizing confirmation, security, picture
film taking care of, cerebrum science, PC correspondence, incitement system, clever card,
law approval, observation et cetera.

A general issue of face affirmation ought to be conceivable in 2 detail both a still image
and clip photo of a scenario. They have been classified into 2 basic applications:
Identification and confirmation. In the distinguishing proof issue, the face to be perceived

1
is obscure and coordinated against a face of a database containing known people.In the
confirmation issue, the framework either affirms or rejects the asserted personality of the
information confront. Essentially, before acknowledgment is played out the framework
should make sense of whether there is a face in a given picture or given video or not. This
procedure is called confront identification. When identification is done, the face locale
ought to be removed or secluded from the foundation scene for the last advance i.e.face
acknowledgment.

The idea of facial acknowledgment is as old as the vision of PC and in light of the useful
significance of the subject and hypothetical enthusiasm from intellectual science. Face
acknowledgment isn't the main technique for perceiving other individuals. Indeed, even
people between each other utilize faculties all together perceive each other.

Since the beginning of this field of innovation there were two principle approaches for
confront acknowledgment:

• Geometrical approach

• Pictorial approach

The geometrical approach utilizes the spatial arrangement of facial highlights while on
the opposite side the other approach i.e. the pictorial approach utilizes formats for facial
highlights. Aside from these two methodologies, there are numerous different
methodologies that have been utilized and that will be utilized as a part of this

undertaking which we will talk about in later sections.

1.2 PROBLEM STATEMENT

Face recognition in itself is an especially complex work which includes area,
identification, and acknowledgment of countenances in any foundation took after by
standardization and acknowledgment. The human face in itself is an extremely muddled
example to identify and perceive, on the grounds that while its life structures is
sufficiently unbending so all countenances have a similar structure, in the meantime there

2
are a great deal of natural and individual variables influencing facial appearance. The rule
issue of face affirmation is broad variability of the recorded pictures as a result of stance,
light conditions, outward appearances, usage of beautifiers, changed haircut, the
proximity of glasses, .Images of a comparative individual taken at different conditions
may now and again indicate more prominent capriciousness because of the already said
factors (intrapersonal vacillation) than pictures of different individuals on account of sex
introduction, race, age and individual assortments (extra individual irregularity)." One
technique for adjusting to intrapersonal assortments is joining into the planning set
pictures with such assortments." And while this is an average practice for assortments, for
instance, outward appearances, use of improving operators and closeness of glasses or
facial hair, it may not be productive if there ought to emerge an event of lighting up or
position assortments. Another critical parameter in stand up to affirmation is developing.
A capable affirmation structure should have the ability to see an individual even
following a couple of years, especially in mug-shot planning quantifiable applications.
This is a to a great degree troublesome endeavor, which has not been viably tended to yet.

1.3 OBJECTIVES
The main motto of our project is to build a software artifact which can detect faces from
the input pictures and then can recognize the face from the picture by matching it with the
database.

The implementation of face recognition technology includes the following three stages:

1. Acquiring data
2. Processing the input
3. Classifying the image and then making a decision

3
Acquiring data:

A simple recorded video of the speaker or any still picture can be considered as input
data. Gathering a 25 outline video gathering requires a case of 1 sec length. To get a 3D
depection of the face one or more than one cameras can be utilized.

Processing the input:

Lighting and changes in shading conditions are taken care by an already existing module
which basically finds the eye position. A face as in input must be perceived to start with
the closeness of appearances. Institutionalization process may be required to bring the
estimations of the live facial case once the face is recognized.

Classification of face images and decision making

Some facial affirmation approaches use the whole face while others center around facial
sections and also regions. The nearness of the face can change on account of outward
appearances.

1.4 METHODOLOGY
There are two major approaches that are used for face detection and recognition, namely

pictorial and geometric based approaches. Apart from these two approaches, there are
many other approaches that have been used and that will be used in this project which we
will discuss in later.

1.5 ORGANISATION
The organization of the report has been made such that the project report has been
segmented into 5 parts: Details about the introduction part are provided in Part 1.The
various methodologies used has been discussed in Part 2 .The various stages that have
been used for the development has been discussed in Part 3.The main final codes that
have been used are placed accordingly in Part 4.The drawbacks , the outcomes the
conclusion etc along with the future scope has been discussed in Part 5.

4
 CHAPTER -2
LITERATURE SURVEY

This project of ours on facial recognition has helped us to have a complete detailed
survey of a number of facial recognition algos along with their pros and cons Some of the
important methods taken into consideration and referred are listed below. The system
basically consists of the following tasks:
1. Acquiring the face
2. Feature extraction
3. Recognition.

2.1 Face Detection Approaches

A still image or a video sequence can be used for “Face recognition”. Discussed below
are some basic “face detection” methods :
1) Methods based on knowledge are basically dependent on the researcher’s know how of
human facial features.
2) Methods based on features include the Invariant features of faces that are used for
detecting texture, skin color etc.
3) Template matching: The image used as an Input is compared with predefined face
templates already present.The disadvantage is the change is pose, color and shape.
4) Appearance-based strategy: In format coordinating strategies, the layouts are
predefined by specialists. Measurable investigation and machine learning methods can be
utilized to locate the significant attributes of face and non-confront pictures.

2.2 Face Recognition Approaches

Various features such as eyes, nose etc are the basic terms on which the analysis of the
structure of the face is done using LFA method of recognition. This system offers better
strength against other fundamental techniques on the facial picture in doing a match.
Nevertheless, with a particular true objective to win in a practical set-up, the cases ought
to be satisfactorily tremendous in number to speak to assortments. Model Matching

5
techniques for face recognition prepare a model for each individual amid model learning
and pick the best coordinating model .Here additionally a major sensible agent model is
vital for good outcomes. Here furthermore a noteworthy sensible operator show is crucial
for good results. An affirmation system in perspective of meager depiction enlisted by
minimization works with the basic idea of giving the affirmation a part as an insufficient
depiction issue. The presence of a large number of features result to the main primary
disadvantage of this technique. It is an energetic and versatile count for confront
affirmation in perspective of straight or arched programming.

Figure 2.1 Generic face recognition configuration

6
Some more “face recognition” techniques are discussed below:

1. Holistic Approach
The complete face region is considered as input to the system in “Holistic approach” for
face detection framework. Examples of comprehensive strategies are “Eigenfaces”,
“Probabilistic eigenfaces”, “Fisherfaces” are the supporters of NFL i.e. (Nearest feature
lines).

2. Hybrid Approach
The technique is made conceivable from how human vision framework sees both
comprehensive and neighbourhood component. The fundamental perspectives that effect
the execution of mixture approach consolidate how to make sense of which highlights are
to be joined and the way they are to join, with a specific end goal to ensure their points of
interest and dismiss their negative marks meanwhile. These issues have cozy association
with the different classifier framework.. In any case, even in such fields, such issues stay
unresolved. Still various endeavours are being made in such field to give us a few
experiences into taking care of these issues, and these lessons can be utilized as help lines
in planning a cross breed face detection framework. The table portrays that
neighbourhood highlights & the ones worldwide are freely fragile to various assortment
factors.

7
 “Variation “Local “Holistic
Factors” Features” Features”

Small NS S

Large S VS

Illuminations VS S

Expressions NS S

Pose S VS

Noise VS S

Occlusion NS VS

Table 1: Difference between two types of feature

2.3 Review of Model Methods:

1. Hidden Markov Models Method

For frontal human face pictures, vital facial highlights show up in a best to howl path, for
example, hair, brow, eyes, nose, mouth, and jaw. This"Hidden Markov models(HMM)" is
someother technique which operates fine on pictures using variety in light changes,
outward appearance & introduction. Gee, an arrangement of accountable models which
portray features of signs. It has great execution in discourse & character acknowledgment.
Framework being displayed is thought to be a Markov procedure with boundaries not
known & objective is to discover concealed boundaries from the recognizable

8
boundaries. Every stage in HMM has a likelihood circulation over conceivable yield
while every stage in a general Markov demonstrate is observable."The creator takes the
assistance of a system called DCT compression property for removing highlight ."A
picture is being isolated into parts of a sub-picture which is additionally connected with
perception vector.

In HHM, there are Markov chain that cannot be observed using a small number of
status in that model. The symbol for observation probability matrix B, a state changing
probability matrix A, starting state distribution π, and a group of probabilitydensity
functions (PDF).Definition of HMM: λ= (A, B, π).

Allocation Of each facial region is done to one state in 1-D.

Figure 2.2: For face recognition

Each face picture with a with "W" and a stature "H" is additionally separated into
covering pieces of tallness "L" and the width being the same.

The sum by which covering "P" importantly affects acknowledgment rate since highlights
are captured autonomous of the vertical position. The estimation of "L" is likewise
urgent.

According to the Hidden Markov model: "The little length "L" will appoint it data which
isn't sufficient to separate to the perception vector. Then again, vast estimation of L will
improve the likelihood of cutting over the feature. .Hence to get a decent better than
average estimation of “L" is very imperative At the moment that the pieces are expelled
from the photograph, a course of action of "DCT" coefficients are determined for each

9
 and every one of them. At the point when each square is changed with "DCT", the most
basic coefficients with low frequencies are met and gathered in a little area in the
"DCT"domain. The degree of recognition vector is really made less of significance,
which along these lines enhances the system beneficial and remembering that up 'til now
having the ability to keep up a good acknowledgment rate. In the readiness arrange, the
photo is isolated all the way where each piece analyzes to a state and starting discernment
probability lattice “B” is gotten from recognition vectors related with each state. Once
"B" is gained, the basic estimation of "An" and "π" are set given the left to right structure
of the face."

Figure 2.3 : Block extraction from image

2. “Neural Network Method”

The know-how of machine learning and various techniques from machine learning are the
basis of “Neural networks:-based methodologies”. The networks can be prepared to catch
more information about the variety of face designs is the basic advantage of utilizing
NNM. The systems must be broadly tuned to get excellent performances the principal
downside of this procedure. Back propagation algorithm has been for the most parts for
this procedure while approaching “face recognition”. However, the meeting of the MLP
systems is slow and the worldwide minima of the error space may not be always
achieved. Quick learning ability and best approximation property are the main features of
RBF neural system on the other side. Along these lines, as of late, numerous specialists
have utilized RBF systems for confront acknowledgment and show in figure 4.

10
Considering all the cases their prosperity percentages are not all that promising as the
blunder percentage which changes from {5 – 9}% under variety of stance, introduction,
and lighting. So this might be because of the choice of the focuses of the hidden layer
neurons that won't have been finished by catching the learning about the conveyance of
preparing examples &varieties of face stance, introduction, and luminosity.

Figure 2.4: RBF Neural Network structure

3. The local binary pattern histogram

We know that “Eigen faces” and “Fisher faces” are both affected by light. In real world
example it if quite tedious to get a constant perfect lighting all time.” LBPH”face
recognizer is an improvement to overcome this drawback.

The main concept behind “LBPH”is not to consider the image as a complete image but
instead, try to find the structure i.e. local by comparison of single pixel with neighbouring
ones.

Consider a “3×3” window and make it move slowly throughout one image. At each
movement, do comparison of the pixel at the center and with its surrounding pixels.

11
Once we are done reading these” 0/1” values of 3x3 matrix in a clockwise manner we will
have a binary pattern like “11100011” which is unique to a specific area image and once
we are done finishing this we will be left with a list of many local binary patterns.

The procedure explained above helps us to get a list of local binary patterns .Once we are
able to fetch those patterns, a conversion takes place which converts each one of the local
binary pattern to a decimal number. The value of the decimal numbers generated are then
further used to make histograms of all the decimal values.

12
 CHAPTER -3
SYSTEM DEVELOPMENTS

In this segment the various choices made during the development time and the
implementation of the many methods used or that will be used during the project are
included.

3.1 DESIGN

In process of development of software for the project various decisions regarding the
nature, design etc of the project have to be made; these include the assumptions made,
APIs used etc. This section discusses these aspects of the project.

3.1.1 ASSUMPTIONS
“PCA” (Principal Component Analysis) is the basis on which various “facial
recognition” algorithms are based. The detected face is compared and analyzed with the
trained image a large number of times by these algorithms, this is their basic principle.
To further improve the accuracy of recognition we are supposed add several images of the
same individual from different angles, varied positions and light conditions.

Method 2:
A study has uncovered that different methodologies and blend of these procedures can be
associated being created of another face acknowledgment structure. Among the different
possible philosophies, we have used a mix of learning based procedures for stand up to
revelation part and neural framework approach for go up against affirmation part.

13
 Figure 3.1: Flow of face detection and recognition

Input Part

For face acknowledgment framework the input of information part is quite essential. The
main activity performed in this part is –“Picture procurement”. For performing picture
handling calculations the live caught pictures are transformed to advance information.
Face identification calculator receives these caught picture information.

Face Detection Part

Face acknowledgment framework requires the removal of face picture task and this is
performed by face identification. in Figure 3 portrays the face recognition part

14
calculation. This tests uncovers that skin division, as an initial step for face identification,
lessens computational time for seeking entire picture.

The division of skin is basically an association, an acknowledgement of the part lying

under that. Depiction of skin which can be used as “RGB” shading space .The picture as
a result of extended light conditions are a result of white modification. Administration of
the white part of the picture should be revised. Results of division:

Figure 3.2: Example of taken/white balance corrected image and skin color segmentation

On the success of the connection of & operation the basic tasks linked with some kind of
skin hopefully seek confront. Boisterous like little locales end, then applied face
applications are selected with 2 conditions. Coverage of few spots and proportions of
bouncing box etc should lie among “0.3”and “1.5”.

Figure 3.3: “Results of Segmentation on Uncorrected (Left) and Corrected Image

(Right)”

15
With behaviour such as these the system goes up against contenders which are confined
from an astounding jumping box input pictures with adjusted input pictures The tallness
of ricocheting box adjusted as 1.28x more noticeable than the breadth of a bobbing box
since parts of the torso will be shed if contender goes along with them These change more
or likely have already been settled.

There is a relation between assertion of contender and the extraction of face picture. The
most critical highlights of face places a special emphasis on the Facial section. eyebrows,
eyes, mouth, nose etc constitute the facial highlights. Evacuation of the eyes and mouth,
two eyes and mouth make isosceles triangle, and division between eye to eye and mid
purpose behind eyes parcel to mouth is corresponding and makes use of this property.
“Laplacian of Gaussian (LoG)” channel and some other segregating assignments are
performed to evacuate facial part of face confident.

Figure 3.4: “Outcome of the option filtering

Figure 3.5: “Facial Feature Extractions (Left) and Face Image (Right) for the author”

16
The analysing segment is done, & pictures are seen in the obtained pictures. “MATLAB”
software is used to execute calculation.(it should likewise be possible in c# which will be
talked about later) and tried for in excess of hundred pictures. This calculation identifies
one face as well as in excess of one face. Little measure of arranged face are satisfactory.
Results are acceptable for general use.

Face Recognition Part

Altered face picture which is acquired in the Face acknowledgment framework, ought to
be grouped to recognize the individual in the database. Face acknowledgment part is
made out of pre-preparing face picture, vector punch picture network, database age, and
after that characterization. The grouping is accomplished by utilizing Feed Forward
Neural Network. .

Figure 3.6 “Algorithm of Face Recognition Part”

In classifier, “Feed Forward Neural Network (FFNN)” is used. FFNN i.e. sustain
forward neural framework is the most central segment of the neural network. Such kind

17
of framework structure is basically used for outline affirmation. Structure points of
interest are: input layer has 900 wellsprings of information.

After the structure is created, by then network ought to be with respect to confronting
database. .Therefore, 900-by-104 size network will get ready grid. Preparing structure
vector part is sorted out with four social affairs in light of the quantity of tests for each
person. Preparing system's portions are produced using a pre-taking care of picture and a
short time later vectorizing to stand up to picture which creates a database.

Method 3 (Final method used)

The local binary pattern histogram

We know that “Eigenfaces” and “Fisherfaces” are bothaffected by light.In real world
example it if quite tedious to get a constant perfect lighting all time.” LBPH”face
recognizer is an improvement to overcome this drawback.The main concept behind
“LBPH”is not to consider the image as a complete image but instead, try to find the
structure ie local by comparison of single pixel with neighbouring ones.Consider a”3×3”
window and make it move slowly throughout one image. At each movement, do
comparison of the pixel at the center and with its surrounding pixels.

Give the neighbours the intensity value lesser than or equal to the main center pixel by 1
and the others 0.

Once we are done reading these” 0/1” values of 3x3 matrix in a clockwise manner we will
have a binary pattern like “11100011” which is unique to a specific area image and once
we are done finishing this we will be left with a list of many local binary patterns.

18
.

Figure 3.7 Binary to decimal

Once we get a list of local binary patterns that are generated from the above mentioned
procedure, we convert each one of them into a decimal number and then make a
histogram of all of those decimal values.

Figure 3.8: Histogram

In the training data set there will be one histogram as a result of this process for each face.
Keeping a track of which histogram belongs to which individual? In the training data set
is the job of this algorithm.

HAAR CLASSIFIER

According to Paul Viola and Michael Jones, “The Haar classifier is a machine learning
based approach.”

It starts by extracting Haar features from each image as shown by the windows below:

19
 A

Figure 3.9 Features

feature is basically calculated by the window i.e. a specific window allocated on the
image. The difference between the sum of pixels under the white part and the sum of
pixels under the black part gives a single value to the feature.

Figure 3.10: Classifiers

20
For the calculation of the features that are in plenty of number, all the sizes of windows
which are possible are placed on all the locations that are possible.

Extraction of 2 features is done as in example in the image above. The area of the eyes is
of a darker shade than the region of the nose and cheeks is the main focus of the first one.
The eyes are of a darker shade than the bridge of the nose is the second property on which
the features rely. Most of the region in an image is a non-face region. this is a
consideration made by an algorithm.

3.2 EXPERIMENTS & RESULTS

A complete hardware and software system is designed and implemented in the Robot
Vision Laboratory of the Department of Mechatronics Engineering at the Atılım
University. The ultimate goal of the larger project (umbrella project) is to develop a
humanoid robot with a narrower application like Guide robot, Guard robot, Office robot,
etc. The developed system has been tested for many live acquired images and results are
satisfactory for such a pioneering work in the department. Improvements are required for
better performance. System description and possible improvements are discussed in this
chapter.

3.2.1 System Hardware

System has three main hardware parts. They are computer, frame grabber, and camera.
Sony EVI-D100P camera and Imagenation PXC 200A frame grabber from CyberOptics
are used.

3.2.2 System Software

Algorithm of system is implemented on MATLAB R2011a software. Image Acquisition

Toolbox, Image Processing Toolbox, and Neural Network Toolbox are used for algorithm
development.

21
 CHAPTER -4
PERFORMANCE ANALYSIS

4.1 METHODS UTILISED

After all the assumptions made in the earlier chapters for every stage, in this segment we
include various methodologies that we used in each stage along with the final code.

4.1.1 Method 1: Code of capturing the dataset:

22
23
4.1.2 Method 2: Code For Training Dataset

24
25
26
4.1.3 Method 3:

Code for recognizing:

27
 CHAPTER -5

CONCLUSIONS

In this segment all the learning outcomes are discussed further carried forward by the
plans and the management put into the project. The basic conclusion and the drawbacks
are also discussed along with the scope and future work of the project.

5.1 REFLECTION

We mentioned above in the “context section” that we have considered various methods
and “strategies” for “face detection and recognition”. Some assumptions were made
initially and we stuck to them as they helped us in achieving our objective. We selected 3
methods for every stage while excluding the “recognition” one and in the expectation of
result. We selected one of them or fusion of two. We actually learnt various strategies to
remove essential data from the image. Adding to this we also learnt about difference in
shading spaces and its different applications. In the end we learnt about “open CV” and
the various opportunities it gives.

5.2 PROJECT CONCLUSION

The face detection and recognition calculations were altogether examined taking various
test pictures and changing the conditions and factors. All the work specified above
included continuous information. The PCA achievement rates were given while for
confront discovery, the achievement rate was diverse for different pictures relying upon
the outer elements. The general achievement rate was 95%.

5.2.1 LIMITATIONS

As discussed above, a few assumptions were made for the project, but the system has few
limitations in spite of that. Even though HD video is quite low in resolution when
compared with digital camera images, it still occupies significant amounts of4disk

28
space. The limitations for HAAR is that it takes more time for training and i s less
accurate for black faces. It is computationally complex and slow.

5.3 FUTURE SCOPE

Further, we may likewise fabricate a GUI in view of python as well as Open CV. Face
acknowledgment frameworks are a piece of facial picture preparing applications and their
noteworthiness as an examination territory are expanding as of late. Usage of framework
are wrongdoing avoidance, video observation, individual check, and comparable security
exercises.

29
 REFERENCES

1. Kim H.PH., Survey paper: FaceKDetection and Face Recognition.LDepartment of

Computer science,KUniversityof Saskatchewan,KSouth Korea, 2004, 1-7.
2. Zhao W., Chellappa R., Phillips P. J., and Rosenfeld A., Face recognition: A
literature survey, ACM Computing Surverys (CSUR), 35(4), 2003, 399-458.
3. Rowley H., Baluja S., and Kanade T., Neural network-based face detection, IEEE
Trans. PatternAnal. Mach. Intell, 20, 1998
4. RanganathS., and arun K., FaceTTrecognition usingKtransform features and
neural networks, PatternRecognition
5. Bellheumur P. N., Hezpanha J. P., and Kriegeaman D. J., Eigenfaces vs.
Fisherfaces: Recognition Using Class Specific Linear Projection, IEEE
Transactionson pattern analysis and machine intelligence
6. Howell, H. Buxton, Learning identity with radial basis function networks,
Neurocomputing, l 20, 1998.
7. Kelly M. D., Visual Identification of People by Computer. Stanford AI Project,
Stanford, CA, Technical Report, 1970
8. Neffian A.V., and el at., Hidden Markov models for face recognition, In
Proceedings, InternationalConference on Acoustics, Speech and Signal
Proceeding,
9. Duda R., and Hart P., Pattern Classification and Scene Analysis. New York:
Wiley, 1973.
10. A. Reda and Dr BoukheilifAoued, 2002, “Artificial Neural Network-Based Face
Recognition

30