Page 1
A Project Report on
Face Recognition using Principle Component Analysis algorithm
Submitted By
Priyanshu chaurasia
Vinay kr. singh
Abhishek bajpayee
Kuldeep kumar
Santosh yadav
DEPARTMENT OF ELECTRONICS & COMMUNICATION ENGINEERING
VIDYA BHAVAN COLLEGE FOR ENGINEERING TECHNOLOGY, KANPUR
UTTAR PRADESH TECHNICAL UNIVERSITY, LUCKNOW
Academic Session - 2015
Page 2
A Project Report on
Face Recognition using Principle Component Analysis algorithm
SUBMIITTED BY
Priyanshu chaurasia
Vinay kr. Singh
Abhishek Bajpayee
Kuldeep kumar
Santosh yadav
submitted in partial fulfillment of therequirements for the award of the degree
of
Bachelor of Technology
IN
ELECTRONICS & COMMUNICATION ENGINEERING
VIDYA BHAVAN COLLEGE FOR ENGINEERING TECHNOLOGY, KANPUR
UTTAR PRADESH TECHNICAL UNIVERSITY, LUCKNOW
Academic Session- 2015
Page 3
ii
CERTIFICATE
I hereby certify that the work which is being presented in the B.Tech. Major Project Report
entitled “Face Recognition using Principle Component Analysis algorithm”, in partial
fulfillment of the requirements for the award of the Bachelor of Technology in Electronics
& Communication Engineering and submitted to the Department of Electronics &
Communication Engineering of Vidya Bhavan College of Engineering, Kanpur is an
authentic record of my own work carried out during a period from January 2015 to June
2015 under the supervision of Uttam yadav, Hod E&C Department.
The matter presented in this thesis has not been submitted by me for the award of any
other degree elsewhere.
This is to certify that the above statement made by the candidate is correct to the best
of my knowledge.
Signature of Supervisor(s)
DEPARTMENT OF ELECTRONICS AND COMMUNICATION,
VIDYA BAHVAN COLLEGE FOR ENGINEERING AND TECHNOLOGY, KANPUR
Page 4
iii
Candidate’s Declaration
I hereby declare that the work presented in this dissertation entitled “Face
Recognition Using Principal Component Analysis Algorithm” in partial
fulfillment for the award of degree of Bachelor of Technology submitted in the
department of Electronics and Communication Engineering, Vidya Bhavan college
for Engineering Technology, Kanpur(Affilited Uttar pradesh Technical University,
Lucknow) is my own work carried out, under the guidance of Mr. Uttam Yadav,
Professor in the department Electronics and Communication Engineering, Vidya
Bhavan College For Engineering and Technology, Kanpur, U.P.
Date: Priyanshu Chaurasia(1141831042)
Vinay Kr. Singh(1141831054)
Abhisek Bajpayee(1141831001)
Kuldeep Kumar(1141831028)
Santosh Yadav(1041831021)
This is to certify that the above statements made by the candidate are correct to the
best of my knowledge.
Date:
(Mr. Uttam Yadav) (Mr. Uttam Yadav) Reader & Head Professor
Department of Department of E&C Engineering
E & C Engineering V.B.C.E.T., Kanpur (U.P.)
V.B.C.E.T., Kanpur (U.P.)
Page 5
iv
ACKNOWLEDGEMENT
I would like to extend my gratitude and my sincere thanks to my honorable, esteemed
supervisor Uttam yadav, Professor, department of Electronics and Communication
Engineering, V.B.C.E.T., Kanpur for his immeasurable guidance and valuable time that he
devoted for project. I sincerely thank for his exemplary guidance and encouragement. His
trust and support inspired me in the most important moments of making right decisions and I
am glad to work with him.
I would also like to thanks Uttam yadav, Head, department of Electronics and
Communication Engineering, V.B.C.E,T, Kanpur and faculty members Shivam singh, Divya
pandey, Ritesh yadav for their guidance and help. I am thankful to scores of books and
technical papers, from where various ideas were drawn but origin of which is now obscure.
NAME OF STUDENT
Priyanshu chaurasia (1141831042)
Vinay kr.singh(1141831054)
Abhishek bajpayee (1141831001)
Kuldeep kumar (1141831028)
Santosh kumar (1041831021 )
Page 6
v
TABLE OF CONTENTS
1. Chapter 1- Introduction 1-8
1.1- Use the Face for Recognition
1.2- Applications
1.3- General Difficulties
5
6
7
2. Chapter-2 Literature Survey of Face Recognition Techniques 9-25
2.0- face recognition techniques
2.1- Face Recognition from Intensity Images
2.1.1- Featured-based
2.1.2- Advantages and Disadvantages
2.1.3- Holistic
2.1.3.1- Statistical
2.2- Motivation
2.3- Approach
2.4- Ethics and Societal Implications
2.5- Face recognition
2.6- Eigenface-based Recognition
2.7- How Humans Perform Face Recognition
2.8- Face Recognition From a Law Enforcement Perspective
2.9- Current Uses of Face Recognition
2.10- Face recognition techniques and results
2.11- Face Recognition
9
9
9
13
14
14
16
18
18
19
19
20
22
24
24
25
3. Chapter 3 26-31
3.1- Introduction of PCA technology
3.2- Face recognition by using some other algorithms
3.3- Principal Components Analysis
3.4- Theory of PCA
26
27
27
29
Page 7
vi
3.5- Principal component Analysis using matrix
3.6- Limitations of PCA
29
31
4. RESULTS 32-43
5. APPENDIX-A 44-47
6. REFERENCES 48-57
Page 8
vii
LIST OF FIGURE
1. Chapter 1
1.1 Example of face recognition
1.2 Biometric Market Technology
1-8
3
4
2.
chapter2
2.1.1.Geometrical features(white) used in the face recognition
experiments.
2.1.2. 35 manually identified facial features.
2.1.3. Grids for face recognition
2.1.4.a) The twelve fiducial points of interest for face recognition; b)
Feature vector has 21 components; ten distances D1-D10 (normalized
with /(D4+D5)) and eleven profile arcs A1-A11 (normalized with
/(A5+A6)) .
2.1.5. Overall Project Structure.
2.3. Face Identification System
2.7. 1. Staring at the faces in the green circles will cause one to
misidentify the central face with the faces circled in red. This is an
example of face aftereffects .
2.7.2. Photograph during the recording of “We Are the World.”
2.8 :- Figure depicts increasingly controlled environments from left to
right.
9-25
11
11
12
13
16
18
21
21
23
3. Chapter 3
3.1.The Flow diagram for the face recognition.
26-31
26
Page 9
viii
4. Chapter 4
Case1-
4.1. Training Set (AT&T).
4.2 Normalized Training Set (AT&T).
4.3. Mean image.
4.4. Eigen faces.
4.5. Input and Re-constructed images.
4.6. Weight of input image and Euclidian distance of input image.
Case-2
4.7. Input and Re-constructed image with input image absent in
training database.
Case-3
4.8 . Training database images.
4.9. Normalized Training database images.
4.10. The Mean image.
4.11. Eigen face images.
4.12. Input and Re-constructed images.
4.13. Weight of input image and Euclidian distance of input image.
Case-4
4.14. Training database images.
4.15. Normalized Training database images.
4.16. The Mean image.
4.17. Input and Re-constructed images.
4.18. Input and Re-constructed images.
32-43
32
32
33
33
34
34
35
37
37
38
38
38
39
39
40
40
41
41
41
42
42
43
Page 10
1
CHAPTER-1
1. INTRODUCTION
“Biometrics consist of technologies that support automatic identification or verification of
identity based on behavioral or physical traits” [1]. Biometrics validates identities by
calculating unique individual characteristics. The most predominant areas of biometrics
involve fingerprints, iris and facial characteristics, hand geometry, retina, voice and touch.
Over the last twenty years or so there has not been a significant market for Biometric
technologies. They have been consigned to infrequent use in films and in some high-security
government or military installations. Nowadays, Biometrics is increasing it‟s stronghold in
many aspects of both public and private life. For example, in some cases in the computer
industry, Biometrics is replacing the more conventional personal identification numbers
(PIN) and passwords. Although Password/Pin systems and Token systems are still the most
common person verification and identification methods, trouble with forgery, theft and
lapses in users‟ memory pose a very real threat to high security environments which are now
turning to biometric technologies to alleviate this potentially dangerous threat. The areas in
which Biometrics are gaining the most support is in the protection of restricted areas, both
commercial and domestic.
This project is based on the face recognition method of Biometric authentication. Although
other Biometric methods such as fingerprints and iris scans are more accurate methods of
identification; facial recognition provides an ingrained human backup because we
instinctively recognize one another. Face recognition is the process in which the facial
features of someone are recognized and then matched to one of the many faces in a database.
“The intuitive way to do face recognition is to look at the major features of the face and
compare them to the same features on other faces.”[2].
The history of face recognition dates back to the 1960‟s when a semi-automated method was
used to compare facial features. First the key features in the photograph were marked by
hand; key features included eyes, ears, nose and mouth.
Page 11
2
Then the distances and ratios between these marks and a common reference point were
computed and these values were then compared to reference data of other faces. In the early
1970‟s Goldstein, Harmon and Lesk [2] created a face recognition system using 21 particular
markers e.g. hair color and lip thickness. This method was less computerized then the
previous method because many of the measurements had to be made entirely by hand.
The next step in face recognition was made by Fisher and Elshlagerb [2] in the early 1970‟s.
They measured the key features in a face using templates of the features of the different parts
of the face.
They then plotted all the pieces on to a general template. Even though this method was more
automated then the previous it proved to be too inconclusive as the features used did not
include enough distinctive data to represent a face.
Kirby and Sirovich pioneered the eigenface approach in 1988 at Brown University. It was the
first genuinely successful system for automatic recognition of human faces. The system
functions by projecting face images onto a feature space that projects the significant
variations among known face images. These significant features are known as “Eigenfaces”.
This eigenface approach is used in this project.
Fingerprints, hand geometry, iris scans, DNA analysis and to some extent personal signatures
are all biometric identifiers. But the only one that does not delay or interfere with access is
face recognition.
Humans identify others by their face and voice and therefore are likely to be more
comfortable with a system that uses face and voice recognition.
Page 12
3
Figure 1.1 : Example of face recognition [4]
This makes face recognition ideal for high traffic areas which are open to the general public
for e.g. airports and railway stations, ATM‟s, public transportation and businesses of all
kinds. Face recognition gives a record of who was there. Since the record is stored in a
database, known persons can be detected automatically and unknown persons checked
quickly.
Concerns following recent terror attacks and constant threats to safety have created a
pressing need for advanced security. As a result, merchants of sophisticated face recognition
biometrics solutions have found their products very much in demand.
According to the World Face Recognition Biometrics Market, the face recognition market
earned revenues of $186 million in 2005 and is likely to grow at a compound annual growth
rate (CAGR) of 27.5 percent to reach $1021.1 million in 2012 [6].
Page 13
4
Figure 1.2 : Biometric Market Technology [7]
However, to compete effectively with alternative biometric technologies, price reduction of
face biometric solutions is critical. A major competitor that matches the face biometric
system in terms of accuracy and performance is the non-automated fingerprint identification
systems (non-AFIS). A reluctance to lower prices may hamper the adoption of face
recognition biometrics.
The purpose of this project is to create a vision based biometric authentication system for
PCs. Now a days, most PCs use a password based authentication system to determine access
rights. The reason for creating a vision based authentication system is because the password
based system:
Less secure – Anybody may enter anybody else‟s password
Remembering passwords – This may become a problem when a user is accessing a
number of different systems
Tedious – A user has to enter his/her passwords every time the workstation needs to be
locked
Because of the advances in image processing techniques, particularly in the areas of face
detection and face recognition, coupled with the low cost of digital imaging hardware, make
a vision based authentication system quite practical.
Page 14
5
1.1 USE THE FACE FOR RECOGNITION
Biometric-based techniques have emerged as the most promising option for recognizing
individuals in recent years since, instead of authenticating people and granting them access to
physical and virtual domains based on passwords, PINs, smart cards, plastic cards, tokens,
keys and so forth, these methods examine an individual‟s physiological and/or behavioural
characteristics in order to determine and/or ascertain his identity. Passwords and PINs are
hard to remember and can be stolen or guessed; cards, tokens, keys and the like can be
misplaced, forgotten, purloined or duplicated; magnetic cards can become corrupted and
unreadable. However, an individual‟s biological traits cannot be misplaced, forgotten, stolen
or forged.
Biometric-based technologies include identification based on physiological characteristics
(such as face, fingerprints, finger geometry, hand geometry, hand veins, palm, iris, retina, ear
and voice) and behavioural traits (such as gait, signature and keystroke dynamics) [1]. Face
recognition appears to offer several advantages over other biometric methods, a few of which
are outlined here:
Almost all these technologies require some voluntary action by the user, i.e., the user needs
to place his hand on a hand-rest for fingerprinting or hand geometry detection and has to
stand in a fixed position in front of a camera for iris or retina identification. However, face
recognition can be done passively without any explicit action or participation on the part of
the user since face images can be acquired from a distance by a camera. This is particularly
beneficial for security and surveillance purposes. Furthermore, data acquisition in general is
fraught with problems for other biometrics: techniques that rely on hands and fingers can be
rendered useless if the epidermis tissue is damaged in some way (i.e., bruised or cracked).
Iris and retina identification require expensive equipment and are much too sensitive to any
body motion. Voice recognition is susceptible to background noises in public places and
auditory fluctuations on a phone line or tape recording.
Signatures can be modified or forged. However, facial images can be easily obtained with a
couple of inexpensive fixed cameras. Good face recognition algorithms and appropriate pre-
processing of the images can compensate for noise and slight variations in orientation, scale
Page 15
6
and illumination. Finally, technologies that require multiple individuals to use the same
equipment to capture their biological characteristics potentially expose the user to the
transmission of germs and impurities from other users. However, face recognition is totally
non-intrusive and does not carry any such health risks.
1.2 APPLICATION
Face recognition is used for two primary tasks:
1. Verification (one-to-one matching): When presented with a face image of an unknown
individual along with a claim of identity, ascertaining whether the individual is who he/she
claims to be.
2. Identification (one-to-many matching): Given an image of an unknown individual,
determining that person‟s identity by comparing (possibly after encoding) that image with a
database of (possibly encoded) images of known individuals.
There are numerous application areas in which face recognition can be exploited for these
two purposes, a few of which are outlined below.
• Security (access control to buildings, airports/seaports, ATM machines and border
checkpoints [2, 3]; computer/ network security [4]; email authentication on multimedia
workstations).
• Surveillance (a large number of CCTVs can be monitored to look for known criminals,
drug offenders, etc. and authorities can be notified when one is located; for example, this
procedure was used at the Super Bowl 2001 game at Tampa, Florida [5]; in another instance,
according to a CNN report, two cameras linked to state and national databases of sex
offenders, missing children and alleged abductors have been installed recently at Royal Palm
Middle School in Phoenix, Arizona [6]).
• General identity verification (electoral registration, banking, electronic commerce,
identifying newborns, national IDs, passports, drivers‟ licenses, employee IDs).
• Criminal justice systems (mug-shot/booking systems, post-event analysis, forensics).
Page 16
7
• Image database investigations (searching image databases of licensed drivers, benefit
recipients, missing children, immigrants and police bookings).
• “Smart Card” applications (in lieu of maintaining a database of facial images, the face-print
can be stored in a smart card, bar code or magnetic stripe, authentication of which is
performed by matching the live image and the stored template) [7].
• Multi-media environments with adaptive human computer interfaces (part of ubiquitous or
context aware systems, behaviour monitoring at childcare or old people‟s centres,
recognizing a customer and assessing his needs) [8, 9].
• Video indexing (labelling faces in video) [10, 11].
• Witness faces reconstruction [12].
In addition to these applications, the underlying techniques in the current face recognition
technology have also been modified and used for related applications such as gender
classification [13-15], expression recognition [16, 17] and facial feature recognition and
tracking [18]; each of these has its utility in various domains: for instance, expression
recognition can be utilized in the field of medicine for intensive care monitoring [19] while
facial feature recognition and detection can be exploited for tracking a vehicle driver‟s eyes
and thus monitoring his fatigue [20], as well as for stress detection [21].
Face recognition is also being used in conjunction with other biometrics such as speech, iris,
fingerprint, ear and gait recognition in order to enhance the recognition performance of these
methods [8, 22-34].
1.3 GENERAL DIFFICULTY
Face recognition is a specific and hard case of object recognition. The difficulty of this
problem stems from the fact that in their most common form (i.e., the frontal view) faces
appear to be roughly alike and the differences between them are quite subtle. Consequently,
frontal face images form a very dense cluster in image space which makes it virtually
Page 17
8
impossible for traditional pattern recognition techniques to accurately discriminate among
them with a high degree of success [35].
Furthermore, the human face is not a unique, rigid object. Indeed, there are numerous factors
that cause the appearance of the face to vary.
The sources of variation in the facial appearance can be categorized into two groups: intrinsic
factors and extrinsic ones [36].
A) Intrinsic factors are due purely to the physical nature of the face and are independent of
the observer. These factors can be further divided into two classes: intrapersonal and
interpersonal
[37]. Intrapersonal factors are responsible for varying the facial appearance of the same
person, some examples being age, facial expression and facial paraphernalia (facial hair,
glasses, cosmetics, etc.). Interpersonal factors, however, are responsible for the differences in
the facial appearance of different people, some examples being ethnicity and gender.
B) Extrinsic factors cause the appearance of the face to alter via the interaction of light with
the face and the observer. These factors include illumination, pose, scale and imaging
parameters (e.g., resolution, focus, imaging, noise, etc.).
Evaluations of state-of-the-art recognition techniques conducted during the past several
years, such as the FERET evaluations [7, 38], FRVT 2000 [39], FRVT 2002 [40] and the
FAT 2004 [41], have confirmed that age variations, illumination variations and pose
variations are three major problem plaguing current face recognition systems [42].
Although most current face recognition systems work well under constrained conditions (i.e.,
scenarios in which at least a few of the factors contributing to the variability between face
images are controlled), the performance of most of these systems degrades rapidly when they
are put to work under conditions where none of these factors are regulated [43].
Page 18
9
CHAPTER-2
Literature Survey of Face Recognition Techniques
2. Face Recognition Techniques
The method for acquiring face images depends upon the underlying application. For instance,
surveillance applications may best be served by capturing face images by means of a video
camera while image database investigations may require static intensity images taken by a
standard camera.
Some other applications, such as access to top security domains, may even necessitate the
forgoing of the nonintrusive quality of face recognition by requiring the user to stand in front
of a 3D scanner or an infra-red sensor.
Therefore, depending on the face data acquisition methodology, face recognition techniques
can be broadly divided into three categories: methods that operate on intensity images, those
that deal with video sequences, and those that require other sensory data such as 3D
information or infra-red imagery. The following discussion sheds some light on the methods
in each category and attempts to give an idea of some of the benefits and drawbacks of the
schemes mentioned therein in general (for detailed surveys, please see [44, 45]).
2.1- Face Recognition from Intensity Images
Face recognition methods for intensity images fall into two main categories: feature-based
and holistic [46-48]. An overview of some of the well-known methods in these categories is
given below.
2.1.1- Featured-based
Feature-based approaches first process the input image to identify and extract (and measure)
distinctive facial features such as the eyes, mouth, nose, etc., as well as other fiducial marks,
and then compute the geometric relationships among those facial points, thus reducing the
Page 19
10
input facial image to a vector of geometric features. Standard statistical pattern recognition
techniques are then employed to match faces using these measurements.
Early work carried out on automated face recognition was mostly based on these techniques.
One of the earliest such attempts was by Kanade [49], who employed simple image
processing methods to extract a vector of 16 facial parameters - which were ratios of
distances, areas and angles (to compensate for the varying size of the pictures) -and used a
simple Euclidean distance measure for matching to achieve a peak performance of 75% on a
database of 20 different people using 2 images per person (one for reference and one for
testing).
Brunelli and Poggio [46], building upon Kanade‟s approach, computed a vector of 35
geometric features (Fig. 3) from a database of 47 people (4 images per person) and reported a
90% recognition rate. However, they also reported 100% recognition accuracy for the same
database using a simple template-matching approach. More sophisticated feature extraction
techniques involve deformable templates ([50], [51], [52]), Hough transform methods [53],
Reisfeld's symmetry operator [54] and Graf's filtering and morphological operations [55].
However, all of these techniques rely heavily on heuristics such as restricting the search
subspace with geometrical constraints [56]). Furthermore, a certain tolerance must be given
to the models since they can never perfectly fit the structures in the image. However, the use
of a large tolerance value tends to destroy the precision required to recognize individuals on
the basis of the model's final best-fit parameters and makes these techniques insensitive to
the minute variations needed for recognition [37]. More recently, Cox et al. [57] reported a
recognition performance of 95% on a database of 685 images (a single image for each
individual) using a 30-dimensional feature vector derived from 35 facial features (Fig. 2.1.2).
However, the facial features were manually extracted, so it is reasonable to assume that the
recognition performance would have been much lower if an automated, and hence less
precise, feature extraction method had been adopted. In general,current algorithms for
automatic feature extraction do not provide a high degree of accuracy and require
considerable computational capacity [58].
Page 20
11
Fig. 2.1.1: Geometrical features(white) used in the face recognition experiments[46].
Fig. 2.1.2: 35 manually identified facial features [57].
Another well-known feature-based approach is the elastic bunch graph matching method
proposed by Wiskott et al. [59] . This technique is based on Dynamic Link Structures [60]. A
graph for an individual face is generated as follows: a set of fiducial points on the face are
chosen.
Each fiducial point is a node of a full connected graph, and is labeled with the Gabor filters‟
responses applied to a window around the fiducial point. Each arch is labeled with the
distance between the correspondent fiducial points. A representative set of such graphs is
combined into a stack-like structure, called a face bunch graph. Once the system has a face
bunch graph, graphs for new face images can then be generated automatically by Elastic
Bunch Graph Matching. Recognition of a new face image is performed by comparing its
image graph to those of all the known face images and picking the one with the highest
similarity value. Using this architecture, the recognition rate can reach 98% for the first rank
and 99% for the first 10 ranks using a gallery of 250 individuals. The system has been
enhanced to allow it to deal with different poses (Fig. 2.1.3) [61] but the recognition
performance on faces of the same orientation remains the same. Though this method was
Page 21
12
among the best performing ones in the most recent FERET evaluation [62, 63],
it does suffer from the serious drawback of requiring the graph placement for the first 70
faces to be done manually before the elastic graph matching becomes adequately dependable
[64].
Campadelli and Lanzarotti [65] have recently experimented with this technique, where they
have eliminated the need to do the graph placement manually by using parametric models,
based on the deformable templates proposed in [50], to automatically locate fiducial points.
They claim to have obtained the same performances as the elastic bunch graph employed in
[59]. Other recent variations of this approach replace the Gabor features by a graph matching
strategy [66] and HOGs (Histograms of Oriented Gradients [67].
Fig. 2.1.3: Grids for face recognition [61].
Considerable effort has also been devoted to recognizing faces from their profiles [68-72]
since, in this case, feature extraction becomes a somewhat simpler one-dimensional problem
[57, 71]. Kaufman and Breeding [70] reported a recognition rate of 90% using face profiles;
Page 22
13
however, they used a database of only 10 individuals. Harmon et al. [68] obtained
recognition accuracies of 96% on a database of 112 individuals, using a 17-dimensional
feature vector to describe face profiles and utilizing a Euclidean distance measure for
matching. More recently, Liposcak and Loncaric [71] reported a 90% accuracy rate on a
database of 30 individuals, using subspace filtering to derive a 21- dimensional feature vector
to describe the face profiles and employing a Euclidean distance measure to match them (Fig.
2.1.4).
Fig. 2.1.4: a) The twelve fiducial points of interest for face recognition; b) Feature vector has
21 components; ten distances D1-D10 (normalized with /(D4+D5)) and eleven profile arcs
A1-A11 (normalized with /(A5+A6)) [71].
2.1.2- Advantages and Disadvantages
The main advantage offered by the featured-based techniques is that since the extraction of
the feature points precedes the analysis done for matching the image to that of a known
individual, such methods are relatively robust to position variations in the input image [37].
In principle, feature-based schemes can be made invariant to size, orientation and/or lighting
[57]. Other benefits of these schemes include the compactness of representation of the face
images and high speed matching [73].
Page 23
14
The major disadvantage of these approaches is the difficulty of automatic feature detection
(as discussed above) and the fact that the implementer of any of these techniques has to make
arbitrary decisions about which features are important [74]. After all, if the feature set lacks
discrimination ability, no amount of subsequent processing can compensate for that intrinsic
deficiency [57].
2.1.3- Holistic
Holistic approaches attempt to identify faces using global representations, i.e., descriptions
based on the entire image rather than on local features of the face. These schemes can be
subdivided into two groups: statistical and AI approaches. An overview of some of the
methods in these categories follows.
2.1.3.1- Statistical
In the simplest version of the holistic approaches, the image is represented as a 2D array of
intensity values and recognition is performed by direct correlation comparisons between the
input face and all the other faces in the database. Though this approach has been shown to
work [75] under limited circumstances (i.e., equal illumination, scale, pose, etc.), it is
computationally very expensive and suffers from the usual shortcomings of straightforward
correlation-based approaches, such as sensitivity to face orientation, size, variable lighting
conditions, background clutter, and noise [76]. The major hindrance to the directmatching
methods‟ recognition performance is that they attempt to perform classification in a space of
very high dimensionality [76]. To counter this curse of dimensionality, several other schemes
have been proposed that employ statistical dimensionality reduction methods to obtain and
retain the most meaningful feature dimensions before performing recognition. A few of these
are mentioned below.
Sirovich and Kirby [77] were the first to utilize Principal Components Analysis (PCA) [78,
79] to economically represent face images. They demonstrated that any particular face can be
efficiently represented along the eigenpictures coordinate space, and that any face can be
approximately reconstructed by using just a small collection of eigenpictures and the
corresponding projections („coefficients‟) along each eigenpicture.
Page 24
15
Turk and Pentland [80, 81] realized, based on Sirovich and Kirby‟s findings, that projections
along eigenpictures could be used as classification features to recognize faces.
They employed this reasoning to develop a face recognition system that builds eigenfaces,
which correspond to the eigenvectors associated with the dominant eigenvalues of the known
face (patterns) covariance matrix, and then recognizes particular faces by comparing their
projections along the eigenfaces to those of the face images of the known individuals. The
eigenfaces define a feature space that drastically reduces the dimensionality of the original
space, and face identification is carried out in this reduced space. An example training set,
the average face and the top seven eigenfaces derived from the training images are shown in
figure., respectively. The method was tested using a database of 2,500 images of 16 people
under all combinations of 3 head orientations, 3 head sizes or scales, and 3 lighting
conditions and various resolutions.
Recognition rates of 96%, 85% and 64% were reported for lighting, orientation and scale
variation. Though the method appears to be fairly robust to lighting variations, its
performance degrades with scale changes.
The aim of this project is to build a face recognition authentication system that:
Retrieves images from a camera in real-time.
Detects the presence of a face in the image.
Identifies the face against some enrolled images.
Through integration with the PCs authentication system, logs on the user
corresponding to the identified face.
Page 25
16
Camera
Capture Images
Transfer to PC
Face Detection
Detect face in image
Face recognition
Analyse face in image
Recognise the face
Authentication
Authenticate the user
whose face is in the
image
Fig 2.1.5 : Overall Project Structure
The initial plan for this project was:
Get a face detection program up and running.
Implement that program with the authentication system to create an initial working
demo.
Get a face recognition program working.
Detect a users face with the face recognition program and log the user on.
Add a utility to add new users to the database of authorized faces.
Integrate the above programs into a fully functional face recognition authentication
system.
2.2 Motivation
Face detection plays an important role in today‟s world. It has many real-world applications
like human/computer interface, surveillance, authentication and video indexing. However
research in this field is still young. Face recognition depends heavily on the particular choice
Page 26
17
of features used by the classifier. One usually starts with a given set of features and then
attempts to derive a optimal subset (under some criteria) of features leading to high
classification performance with the expectation that similar performance can also be
displayed on future trials using novel (unseen) test data.
Interactive Face Recognition (IFR) can benefit the areas of: Law Enforcement, Airport
Security, Access Control, Driver's Licenses & Passports, Homeland Defense, Customs &
Immigration and Scene Analysis. The following paragraphs detail each of these topics, in
turn
Law Enforcement: Today's law enforcement agencies are looking for innovative
technologies to help them stay one step ahead of the world's ever-advancing terrorists.
Airport Security: IFR can enhance security efforts already underway at most airports and
other major transportation hubs (seaports, train stations, etc.). This includes the identification
of known terrorists before they get onto an airplane or into a secure location.
Access Control: IFR can enhance security efforts considerably. Biometric identification
ensures that a person is who they claim to be, eliminating any worry of someone using
illicitly obtained keys or access cards.
Driver's Licenses & Passports: IFR can leverage the existing identification infrastructure.
This includes, using existing photo databases and the existing enrollment technology (e.g.
cameras and capture stations); and integrate with terrorist watch lists, including regional,
national, and international "most-wanted" databases.
Homeland Defense: IFR can help in the war on terrorism, enhancing security efforts. This
includes scanning passengers at ports of entry; integrating with CCTV cameras for "out-of-
the-ordinary" surveillance of buildings and facilities; and more.
Customs & Immigration: New laws require advanced submission of manifests from planes
and ships arriving from abroad; this should enable the system to assist in identification of
individuals who should, and should not be there.
Page 27
18
2.3 Approach
The basic algorithm starts with a pre-processing step, consisting of digitization and
segmentation. The next step is called face segmentation. We define the face segmentation
problem as: given a scene that may contain one or more faces, create sub-images that crop
out individual faces. After face segmentation, the device enters into the face identification
mode, as shown.
(small)
Suspect database
Face
Data Base
Feature
Data Base
Face
Segmentation
Feature
Extract ionclassifier
matches
GUI
Displays possible
candidat es for selection
Fig 2.3 :- Face Identification System
Human skin is relatively easy to detect in controlled environments, but detection in
uncontrolled settings is still an open problem [6.]. Many approaches to face detection are
only applicable to static images assumed to contain a single face in a particular part of the
image. Additional assumptions are placed on pose, lighting, and facial expression. When
confronted with a scene containing an unknown number of faces, at unknown locations, they
are prone to high false detection rates and computational inefficiency. Real-world images
have many sources of corruption (noise, background activity, and lighting variation) where
objects of interest, such as people, may only appear at low resolution. The problem of
reliably and efficiently detecting human faces is attracting considerable interest. An earlier
generation of such a system has already been used for the purpose of flower identification by
[7, 8].
2.4- Ethics and Societal Implications
Face detection is the fastest growing biometric technology today [2]. Despite their lingering
questions regarding the practical usefulness of facial identification technology,
Page 28
19
law enforcement and military facial identification systems have been in place for several
years without arousing too much controversy. According to industry insiders, this is because
these applications have proven quite successful in carrying out specific objectives and the
public is often unaware of these uses. After September 11, many of the face recognition
companies redoubled its efforts to create reliable facial recognition equipment. According to
study, the industry still has a lot of work to do. Even though there are lots of advantages in a
face recognition system. Some people still feel that face recognition system invades privacy
of a citizen.
Also the accuracy of the systems is of concern. Even if a subject's face is stored in the
database, a disguise or even a minor change in appearance, like wearing sunglasses or
wearing or growing a mustache can often fool the system. Even an unusual facial expression
can confuse the software. Facial identifiers often cannot distinguish twins. Other factors
affecting the reliability of the images are changes in the lighting and the angle at which the
photos are taken. The systems often have difficulty recognizing the effects of aging [35].
2.5- Face Recognition
Face recognition is a biometric which uses computer software to determine the identity of the
individual. Face recognition falls into the category of biometrics which is “the automatic
recognition of a person using distinguishing traits” [6]. Other types of biometrics include
fingerprinting, retina scans, and iris scan.
2.6 Eigenface-based Recognition
2D face recognition using eigenfaces is one of the oldest types of face recognition. Turk and
Pentland published the groundbreaking “Face Recognition Using Eigenfaces” in1991. The
method works by analyzing face images and computing eigenfaces which are faces
composed of eigenvectors. The comparison of eigenfaces is used to identify the presence of a
face and its identity.
There is a five step process involved with the system developed by Turk and Pentland. First,
the system needs to be initialized by feeding it a set of training images of faces. This is used
these to define the face space which is set of images that are face like. Next, when a face is
Page 29
20
encountered it calculates an eigenface for it. By comparing it with known faces and using
some statistical analysis it can be determined whether the image presented is a face at all.
Then, if an image is determined to be a face the system will determine whether it knows the
identity of it or not. The optional final step is that if an unknown face is seen repeatedly, the
system can learn to recognize it.
The eigenface technique is simple, efficient, and yields generally good results in controlled
circumstances [1]. The system was even tested to track faces on film. There are also some
limitations of eigenfaces. There is limited robustness to changes in lighting, angle, and
distance [6]. 2D recognition systems do not capture the actual size of the face, which is a
fundamental problem [4]. These limits affect the technique‟s application with security
cameras because frontal shots and consistent lighting cannot be relied upon.
2.7 How Humans Perform Face Recognition
It is important for researchers to know the results of studies on human face recognition [8].
Knowing these results may help them develop ground breaking new methods. After all,
rivaling and surpassing the ability of humans is the key goal of computer face recognition
research.
The key results of a 2006 paper “Face Recognition by Humans: Nineteen Results All
Computer Vision Researchers Should Know About” are as follows:
1. Humans can recognize familiar faces in very low-resolution images.
2. The ability to tolerate degradations increases with familiarity.
3. High-frequency information by itself is insufficient for good face recognition
performance.
4. Facial features are processed holistically.
5. Of the different facial features, eyebrows are among the most important for
recognition.
6. The important configural relationships appear to be independent across the width and
height dimensions.
Page 30
21
7. Face-shape appears to be encoded in a slightly caricatured manner.
8. Prolonged face viewing can lead to high level aftereffects, which suggest prototype-
based encoding.
Fig. 2.7.1 :- Staring at the faces in the green circles will cause one to misidentify the
central face with the faces circled in red. This is an example of face aftereffects [8].
9. Pigmentation cues are at least as important as shape cues.
10. Color cues play a significant role, especially when shape cues are degraded.
11. Contrast polarity inversion dramatically impairs recognition performance, possibly due
to compromised ability to use pigmentation cues.
Fig.2.7.2 :- Photograph during the recording of “We Are the World.” This figure
demonstrates how polarity inversion effects face recognition in humans.
Page 31
22
Several famous artists are in the picture including Ray Charles, Lionel Ritchie, Stevie
Wonder, Michael Jackson, Tina Turner, Bruce Springstein, and Billy Joel though they
are very difficult to identify.
12. Illumination changes influence generalization.
13. View-generalization appears to be mediated by temporal association.
14. Motion of faces appears to facilitate subsequent recognition.
15. The visual system starts with a rudimentary preference for face-like patterns.
16. The visual system progresses from a piecemeal to a holistic strategy over the first
several year of life.
17. The human visual system appears to devote specialized neural resources for face
perception.
18. Latency of responses to faces in infer temporal (IT) cortex is about 120 ms, suggesting
a largely feed forward computation.
19. Facial identity and expression might be processed by separate systems.
2.8 Face Recognition From a Law Enforcement Perspective
Facial recognition is attractive for law enforcement. It can be used in conjunction with
existing surveillance camera infrastructure to hunt for know criminals. Face recognition is
covert and non intrusive, opposed to other biometrics such as finger prints, retina scans, and
iris scans [6]. This is especially important in conjunction with the law because faces are
considered public. Comprehensive photo databases from mug shots or driver‟s licenses
already exist.
Because of difficulties face recognition has with respect to lighting, angle, and other
factors, it is advantageous to attempt to get as high quality images with regard to these
factors. Facetraps are a concept where cameras are strategically placed in order to obtain
relatively controlled photographs [6]. Examples are placing cameras facing doorways, at
airport check-ins, or near objects people are likely to stare at.
Page 32
23
These traps would aid face recognition software by helping to capture a straight frontal
image which allow for higher accuracy of the system. Despite their potential benefit, there
appears to be very little research done on facetraps.
Fig.2.8 :- Figure depicts increasingly controlled environments from left to right. From left to
right: suspect on a plane (no control), subject at a check-in counter, subject on an escalator
staring at a flashing red bulb, subject passing through a doorway, subject sitting in front of a
camera (perfect control) [6].
Some have questioned the legality of face scanning and have argued that such systems
which are used to hunt to criminals in public places are an invasion of privacy. From a legal
perspective, in the United States, one does not have a right to privacy for things shown in
public [6]. “What a person knowingly exposes to the public. . . is not a subject of Fourth
Amendment protection,” United States v. Miller, 425 U.S. 435 (1976). “No person can have
a reasonable expectation that others will not know the sound of his voice, any more than he
can reasonably expect that his face will be a mystery to the world,” United States v. Dionisio,
410 U.S. 1 (1973). These excerpts from Supreme Court decisions help to establish that face
recognition is constitutional. Face recognition must be improved further before it becomes
Page 33
24
a useful tool for law enforcement. It remains to be seen what the right balance is, socially
speaking, between maximizing public safety and respecting individual rights.
2.9 Current Uses of Face Recognition
Face recognition systems used tied to surveillance cameras in Tampa, Florida and Newham,
Great Britain [2]. Trials of the systems yielded poor results. The New ham system didn‟t
result in a single arrest being made in three years. Logan Airport, in Boston, performed two
trials of face recognition systems. The system achieved only 61.7% accuracy [5].
Australian customs recently rolled out its Smart Gate system to automate checking faces
with passport photos. Google is testing face recognition using a hidden feature in its image
searching website [7]. Google purchased computer vision company Neven Vision in 2006
and plans to implement its technology into its Picasa photo software.
2.10 Face recognition techniques and results
Use of biometrics has increased over last few years due to its inherent advantages over
customary identification tools such as token card and password etc. In biometrics, after
fingerprint, face recognition is second most preferred method with reasonably good accuracy.
In some applications like CCTV cameras where face of a person is available for processing,
face recognition techniques can to be very useful.
A biometric system which relies on a single biometric identifier is most of the time unable to
meet the desired requirements in making a personal identification and verification. This
happens due to the algorithm‟s limitations. Nowadays various biometric identifiers like, face,
finger, voice, palm, retina and hand writing etc. are used. However, each of these methods
has its own advantages and disadvantages. This work proposes a basic face recognition based
system which is fuzzy fusion of principal component analysis (PCA), independent
component analysis (ICA) and linear discriminant analysis (ILDA) algorithms. In this
chapter, Face identification techniques PCA, ICA and ILDA are discussed. The basic
structure of the algorithms is detailed. For the above mentioned algorithms simulation results
are also presented.
Page 34
25
2.11 Face Recognition:
It is quite easy to obtain facial images with a couple of inexpensive fixed cameras. Good face
recognition algorithms and appropriate preprocessing of the images can compensate for noise
and slight variations in orientation, scale and illumination [3].
Face recognition is used for two primary purposes:
1. Verification (one-to-one matching): When presented with a face image of an unknown
individual along with a claim of identity, making sure whether the individual is who he/she
claims to be.
2. Identification (one-to-many matching): Given an image of an unknown individual,
determining the identity of that person by comparing (possibly after encoding) that image
with a database of (possibly encoded) images of known individuals [3].
Page 35
26
Chapter- 3
PCA ANALYSIS
Computational models of faces have been an active area of research since late 1980s, for they
can contribute not only to theoretical insights but also to practical applications, such as
criminal identification, security systems, image and film processing, and human-computer
interaction, etc. However, developing a computational model of face recognition is quite
difficult, because faces are complex, multidimensional, and subject to change over time.
3.1 Introduction
The flow diagram for the face recognition techniques is shown in the figure 12 given below:
Fig.3.1 :- The Flow diagram for the face recognition
In the face recognition system the flow must be followed. It defines all the required steps in
the face recognition system. Figure 3.1 defines all the required steps, but the most important
step is the Feature Extraction which is ultimately used for dimensional reduction as well as
for extracting features from input of the system. Extracted features are passed to the last
phase that is classification where the identification or verification rate is calculated.
Page 36
27
The most popular methods used in the face recognition are:
1. Principal Component Analysis
2. 2.Independent Component Analysis
3. 3.Linear Discriminate Analysis
3.2- Face recognition
The process of face recognition involves the examination of facial features in an image,
recognizing those features and matching them to one of the many faces in the database.
There are many algorithms capable of performing face recognition; such as:
Principal Component Analysis
Discrete Cosine Transform
3D recognition methods
Gabor Wavelets method
Hidden Markov Models
Kernel methods
The Principal Component Analysis (PCA) method of face recognition is used in this
biometric authentication system. In this chapter we will discuss why the PCA method was
chosen, the theory of face recognition using the PCA method and how it is used in this
system.
3.3 Principal Components Analysis
Why use PCA?
There were many issues to consider when choosing a face recognition method. The keys ones
were:
1) Accuracy 3) Process speed
2) Time limitations 4) Availability
Page 37
28
With these in mind the PCA method of face recognition was selected for this project
because:
Simplest and easiest method to implement – due to project deadlines this method
seemed the most practical.
Very fast computation time.
Accurate – this method is definitely not the most accurate of face recognition
algorithms but considering the requirements of this project it was judged to be accurate
enough
PCA is supported within the Open CV library – this was key because it made
integration with the face detection program very easy
There are disadvantages with the PCA method but they were deemed to be inconsequential.
PCA is:
Translation variant – if the image is shifted or tilted then it will not recognise the
face. This flaw doesn‟t affect this system as the camera will always be located upright
and in front of the user.
Scale variant – scaling the images will affect the performance of face recognition.
This was not a big problem in this project as new users are added using the camera at
the PC so the face to be recognised and the faces in the database are captured in the
same way; making them very similar in
size. Also the user will always be relatively close to the camera eliminating the
possibility of a blurry, out of focus face.
Background variant – recognising a face with a different background is difficult. The
face detection program is altered so it not only detects a face but extracts it to a new
image, thus cutting out almost all the background. The residual sections of background
are insignificant since the camera‟s resolution will only keep the face in focus because
of the distance.
Lighting variant – if the light intensity changes then the accuracy of face recognition
drops. This authentication system involves capturing images in front of a PC and
usually the room is lighted when a user is using a PC.
Page 38
29
Taking all this into consideration, the PCA method of face recognition was deemed to be the
most practical for this project.
3.4 Theory of PCA
Principal Component Analysis is a process that extracts the most relevant information
contained in a face and then tries to build a computational model that best describes it.
The basic theory of PCA can be described in the following steps:
1. Eigenvectors or eigenfaces of the covariance matrix are found. This is done by training
a set of face images.
2. These eigenvectors become the eigenspace (a multi-dimensional subspace comprised
of the Eigenvectors) in which every face is projected on.
3. Recognition is performed by comparing the location of a face in the eigenspace with
the location of known users. In other words calculating the Euclidean distance.
This is only the basic theory of PCA. In order to fully understand PCA it needs to be
explained mathematically which is done below.
3.5 Principal component Analysis
A 2-D facial image can be represented as 1-D vector by concatenating each row (or column)
into a long thin vector [4, 5]. Let‟s suppose we have M vectors of size N (= rows of image ×
columns of image) that represents a set of sampled images. jp represents the pixel values.
1 2[ , ... ]T
i Nx p p p 1.......i M (3.1)
The images are mean centered by the subtraction of the mean image from each image vector.
Let 1
1 M
i
i
m xM
represent the mean image, and let iw be defined as mean centered image
i iw x m (3.2)
Page 39
30
Our goal is to find a set of ie ‟s which have the largest possible projection onto each of the
iw ‟s. We wish to find a set of M orthonormal vectors ie for which the quantity
2
1
1 MT
i i n
n
e wM
is maximized with the orthonormality constraint
T
l k lke e (3.3)
It has been shown that the ie ‟s and i ‟s are given by the eigenvectors and eigenvalues of the
covariance matrix TC WW , where W is a matrix composed of the column vectors iw placed
side by side [4]. The size of C is N × N which could be enormous. For example, images of
size 64×64 create the covariance matrix of size 4096×4096. It is not practical to solve for the
eigenvectors of C directly. According to a common theorem in linear algebra, the vectors
ie and scalars¸ i can be obtained by solving for the eigenvectors and eigenvalues of the M×M
matrix TW W . Let id and i be the eigenvectors and eigenvalues of TW W respectively.
T
i i iW Wd d (3.4)
By multiplying left to both sides by W
T
i i iWW Wd W d (3.5)
which means that the first 1M eigenvectors ie and eigenvalues i of TWW are given by
iWd and i respectively. iWd needs to be normalized in order to be equal to ie . Since we only
sum up a finite number of image vectors, M , the rank of the covariance matrix cannot
exceed 1M (The -1 come from the subtraction of the mean vector m).
The eigenvectors corresponding to nonzero eigenvalues of the covariance matrix produce an
orthonormal basis for the subspace within which most image data can be represented with a
small amount of error.
The sorting of eigenvectors is done according to their corresponding eigenvalues from high
to low. The eigenvector associated with the largest eigenvalue is one that reflects the greatest
variance in the image [5]. That is, the smallest eigenvalue is associated with the eigenvector
that finds the least variance.
Page 40
31
They decrease in exponential fashion, meaning that the roughly 90% of the total variance is
contained in the first 5% to 10% of the dimensions. A facial image can be projected onto
M M dimensions by computing
1 2[ , ... ]T
Mv v v (3.6)
where T
i i iv e w . iv is the thi coordinate of the facial image in the new space, which came to be
the principal component. The vectors ie are also images, so called, eigenimages, or
eigenfaces in our case, which was first named by [5-9]. They can be viewed as images and
indeed look like faces. The simplest method for determining which face class provides the
best description of an input facial image is to find the face class k that minimizes the
Euclidean distance
k k (3.7)
where, k is a vector describing the thk face class. If k is less than some predefined
threshold, a face is classified as belonging to the class k.
3.6 Limitations of PCA
The main limitations of the PCA are as follows:
1. The face image should be normalized and frontal-view
2. The system is an auto-associative memory system. It is harmful to be over-fitted.
3. Training is very computationally intensive.
4. It is hard to decide suitable thresholds - It is kind of Art!
5. The suggested methods to deal with unknown faces and non-faces are not good enough to
differentiate them from known faces.
Page 41
32
Chapter- 4
Results
Face images for the test are taken from AT&T data base. The database consists of 430
images. We have selected 12 images, for the demonstration of the algorithm. The files are in
PGM format. Each image is displayed by 92×112 pixels, with 256 grey levels per pixel. The
images are arranged in 12 directories (one for each „subject‟), which have names of the
form sX, where X indicates the subject number (between 1 and 25).
Case 1
In the first case 12 images are taken as training set, each with mean 100 and standard
deviation of 80. In the second step the mean and standard deviation of all images are changed
for normalization. This is done to reduce the error due to lighting conditions and background
Fig.4.1:- Training Set (AT&T)
Page 42
33
The normalized images are shown in figure 4.2, and these images are very much similar to
the images in figure, however when background changes abruptly, the normalization is very
effective.
Fig.4.2:- Normalized Training Set (AT&T)
In the next step, the mean image is generated as shown in figure 4.3. The pixel values of the
images ranges form 0 to 255.
Fig.4.3:- Mean image
Page 43
34
In the next step, co-variance matrix is created, thereafter the Eigen-values are obtained, and
the Eigen values close to zero are dropped and for the left over Eigen values, Eigen vector
are obtained. Finally, after the normalization of Eigen vectors, Eigen faces are calculated
(Figure 4.4).
Fig.4.4:- Eigenfaces
In case of user authentication, template matching is done. In figure 4.5, the input image and
the re-constructed image is shown. The re-constructed image is very much similar to the
input image.
Fig.4.5:- Input and Re-constrcuted images
Page 44
35
The pictures, deviate around the forehead portion, as the hair style of the training images are
distinct. Therefore, as in PCA only principal components are considered, it may not be
possible to exactly detect the input image. In figure 4.6 weight of the input image and
Euclidian distance of the input image are shown. The minimum Euclidian distance is 11965
and maximum Euclidian distance is 15547.
Fig.4.6:- Weight of input image and Euclidian distance of input image
Image Minimum Value Maximum Value
11990 14421
11978 15032
11979 14939
Page 45
36
11987 13907
11988 14780
11965 15547
11994 14305
11986 14307
11980 14432
11989 13453
11993 14204
11993 14386
It is observed form the table that the minimum Euclidian distance is somewhat constant and
remains around 11990. However, maximum Euclidian distance varies from 13453 to 15547.
Moreover Euclidian distance is more for the images which contains more information (like:
spectacles).
Page 46
37
Case 2
Considering the case, where an input image is to be matched with template which is not
present in the training database. It is observed from the figure, that the re-constructed is
different from the input image. This is actually a desired feature because if an image is not
present in database then template matching is not possible.
Fig.4.7:- Input and Re-constructed image with input image absent in training database
The minimum Euclidian distance is 12487 and maximum Euclidian 13927.
Page 47
38
Case 3
In the third case, the image dataset of a single person is taken with various facial expressions.
In the dataset 10 images are considered as shown in figure 4.8. The corresponding
normalized images are showninfig.4.9.
Fig.4.8:- Training database images
Fig.4.9:- Normalized Training database images
Page 48
39
Fig.4.10:- The Mean image
The mean image is shown in figure 4.10 and the Eigen-face image is shown in figure 4.11.
Fig.4.11:- Eigenface images
Page 49
40
Fig.4.12:- Input and Re-constrcuted images
Fig.4.13: - Weight of input image and Euclidian distance of input image
The input and re-constructed images is shown in figure 4.12. The Euclidian Distance is
shown in figure 4.13, with maximum and minimum values are 14439 and 12551
respectively. It is clear from the figure that the facial expressions are also captured by the
PCA algorithm.
Page 50
41
Case 4
In this case, total 10 images of two persons are considered with 6 images of a person and 4
images for another person. However, we have a total of 20 images, 10 each in our master
database. In this experiment, we will try to re-construct an image which is not present in
training dataset. In the re-construction phase, the input images will form left over images of
these two persons.
Fig.4.14:- Training database images
In the dataset 10 images (6+4) are considered as shown in figure 4.14, and the normalized
images are shown in figure 4.15 . The mean image is shown in figure 4.16, while the two re-
constructed images are shown in figure 4.17 and figure 4.18 respectively.
Fig.4.15:- Normalized Training database images
Page 51
42
Fig. 4.16:- The Mean image
Fig.4.17:- Input and Re-constructed images
In figure 30, it is clear that the re-constructed image is not very similar to the input image.
However, these two images have some similarity.
As the input image is not in training dataset, therefore PCA is not able to recognize the
person, however, if the test image form the training set is chosen, then the re-construction is
good as shown in figure 4.18.
Page 52
43
Fig.4.18:- Input and Re-constructed images
(2) Regarding the pattern vector representing a face class, we can make each face class
consist of several pattern vectors, each constructed from a face image of the same
individual under a certain condition, rather than taking the average of these vectors to
represent the face class.
Page 53
44
Appendix-A
Matlab Code
% Face recognition Using PCA
clear all
close all
clc
M=10;
um=100;
ustd=80;
S=[];
figure(1);
for i=1:M
str=strcat(int2str(i),'.pgm');
eval('img=imread(str);');
subplot(ceil(sqrt(M)),ceil(sqrt(M)),i)
imshow(img)
if i==3
title('Training set','fontsize',18)
end
drawnow;
[irow icol]=size(img);
temp=reshape(img',irow*icol,1);
S=[S temp];
end
for i=1:size(S,2)
temp=double(S(:,i));
m=mean(temp);
st=std(temp);
S(:,i)=(temp-m)*ustd/st+um;
end
figure(2);
for i=1:M
str=strcat(int2str(i),'.pgm');
img=reshape(S(:,i),icol,irow);
img=img';
eval('imwrite(img,str)');
Page 54
45
subplot(ceil(sqrt(M)),ceil(sqrt(M)),i)
imshow(img)
drawnow;
if i==3
title('Normalized Training Set','fontsize',18)
end
end
m=mean(S,2);
tmimg=uint8(m);
img=reshape(tmimg,icol,irow);
img=img';
figure(3);
imshow(img);
title('Mean Image','fontsize',18)
dbx=[];
for i=1:M
temp=double(S(:,i));
dbx=[dbx temp];
end
A=dbx';
L=A*A';
[vv dd]=eig(L);
v=[];
d=[];
for i=1:size(vv,2)
if(dd(i,i)>1e-4)
v=[v vv(:,i)];
d=[d dd(i,i)];
end
end
[B index]=sort(d);
ind=zeros(size(index));
dtemp=zeros(size(index));
vtemp=zeros(size(v));
len=length(index);
for i=1:len
dtemp(i)=B(len+1-i);
ind(i)=len+1-index(i);
vtemp(:,ind(i))=v(:,i);
end
d=dtemp;
v=vtemp;
for i=1:size(v,2)
Page 55
46
kk=v(:,i);
temp=sqrt(sum(kk.^2));
v(:,i)=v(:,i)./temp;
end
u=[];
for i=1:size(v,2)
temp=sqrt(d(i));
u=[u (dbx*v(:,i))./temp];
end
for i=1:size(u,2)
kk=u(:,i);
temp=sqrt(sum(kk.^2));
u(:,i)=u(:,i)./temp;
end
figure(4);
for i=1:size(u,2)
img=reshape(u(:,i),icol,irow);
img=img';
img=histeq(img,255);
subplot(ceil(sqrt(M)),ceil(sqrt(M)),i)
imshow(img)
drawnow;
if i==3
title('Eigenfaces','fontsize',18)
end
end
omega = [];
for h=1:size(dbx,2)
WW=[];
for i=1:size(u,2)
t = u(:,i)';
WeightOfImage = dot(t,dbx(:,h)');
WW = [WW; WeightOfImage];
end
omega = [omega WW];
end
InputImage=imread('6.pgm')
figure(5)
subplot(1,2,1)
imshow(InputImage); colormap('gray');title('Input image','fontsize',18)
InImage=reshape(double(InputImage)',irow*icol,1);
temp=InImage;
Page 56
47
me=mean(temp);
st=std(temp);
temp=(temp-me)*ustd/st+um;
NormImage = temp;
Difference = temp-m;
p = [];
aa=size(u,2);
for i = 1:aa
pare = dot(NormImage,u(:,i));
p = [p; pare];
end
ReshapedImage = m + u(:,1:aa)*p;
ReshapedImage = reshape(ReshapedImage,icol,irow);
ReshapedImage = ReshapedImage';
subplot(1,2,2)
imagesc(ReshapedImage); colormap('gray');
title('Reconstructed image','fontsize',18)
InImWeight = [];
for i=1:size(u,2)
t = u(:,i)';
WeightOfInputImage = dot(t,Difference');
InImWeight = [InImWeight; WeightOfInputImage];
end
ll = 1:M;
figure(68)
subplot(1,2,1)
stem(ll,InImWeight)
title('Weight of Input Face','fontsize',14)
e=[];
for i=1:size(omega,2)
q = omega(:,i);
DiffWeight = InImWeight-q;
mag = norm(DiffWeight);
e = [e mag];
end
kk = 1:size(e,2);
subplot(1,2,2)
stem(kk,e)
title('Eucledian distance of input image','fontsize',14)
MaximumValue=max(e)
MinimumValue=min(e)
Page 57
48
References
[1] A. K. Jain, R. Bolle, and S. Pankanti, "Biometrics: Personal Identification in Networked
Security," A. K. Jain, R. Bolle, and S. Pankanti, Eds.: Kluwer Academic Publishers, 1999.
[2] K. Kim, "Intelligent Immigration Control System by Using Passport Recognition and
Face Verification," in International Symposium on Neural Networks. Chongqing, China,
2005, pp.147-156.
[3] J. N. K. Liu, M. Wang, and B. Feng, "iBotGuard: an Internet-based intelligent robot
security system using invariant face recognition against intruder," IEEE Transactions on
Systems Man And Cybernetics Part C-Applications And Reviews, Vol.35, pp.97-105, 2005.
[4] H. Moon, "Biometrics Person Authentication Using Projection-Based Face Recognition
System in
Verification Scenario," in International Conference on Bioinformatics and its Applications.
Hong Kong,
China, 2004, pp.207-213.
[5] D. McCullagh, "Call It Super Bowl Face Scan 1," in Wired Magazine, 2001.
[6] CNN, "Education School face scanner to search for sex offenders." Phoenix, Arizona:
The Associated
Press, 2003.
[7] P. J. Phillips, H. Moon, P. J. Rauss, and S. A. Rizvi, "The FERET Evaluation
Methodology for Face
Recognition Algorithms," IEEE Transactions on Pattern Analysis and Machine Intelligence,
Vol.22,
pp.1090-1104, 2000.
[8] T. Choudhry, B. Clarkson, T. Jebara, and A. Pentland, "Multimodal person recognition
using unconstrained audio and video," in Proceedings, International Conference on Audio
and Video-Based Person Authentication, 1999, pp.176-181.
[9] S. L. Wijaya, M. Savvides, and B. V. K. V. Kumar, "Illumination-tolerant face
verification of low-bitrate JPEG2000 wavelet images with advanced correlation filters for
handheld devices," Applied
Page 58
49
Optics, Vol.44, pp.655-665, 2005.
[10] E. Acosta, L. Torres, A. Albiol, and E. J. Delp, "An automatic face detection and
recognition system for video indexing applications," in Proceedings of the IEEE
International Conference on Acoustics, Speech and Signal Processing, Vol.4. Orlando,
Florida, 2002, pp.3644-3647.
[11] J.-H. Lee and W.-Y. Kim, "Video Summarization and Retrieval System Using Face
Recognition and
MPEG-7 Descriptors," in Image and Video Retrieval, Vol.3115, Lecture Notes in Computer
Science:
Springer Berlin / Heidelberg, 2004, pp.179-188.
[12] C. G. Tredoux, Y. Rosenthal, L. d. Costa, and D. Nunez, "Face reconstruction using a
configural,
eigenface-based composite system," in 3rd Biennial Meeting of the Society for Applied
Research in
Memory and Cognition (SARMAC). Boulder, Colorado, USA, 1999.
[13] K. Balci and V. Atalay, "PCA for Gender Estimation: Which Eigenvectors Contribute?"
in Proceedings of Sixteenth International Conference on Pattern Recognition, Vol.3. Quebec
City, Canada, 2002, pp. 363-366.
[14] B. Moghaddam and M. H. Yang, "Learning Gender with Support Faces," IEEE
Transactions on Pattern Analysis and Machine Intelligence, Vol.24, pp.707-711, 2002.
[15] R. Brunelli and T. Poggio, "HyperBF Networks for Gender Classification," Proceedings
of DARPA
Image Understanding Workshop, pp.311-314, 1992.
[16] A. Colmenarez, B. J. Frey, and T. S. Huang, "A probabilistic framework for embedded
face and
facial expression recognition," in Proceedings of the IEEE Conference on Computer Vision
and Pattern
Recognition, Vol.1. Ft. Collins, CO, USA, 1999, pp. 1592-1597.
[17] Y. Shinohara and N. Otsu, "Facial Expression Recognition Using Fisher Weight Maps,"
in Sixth
Page 59
50
IEEE International Conference on Automatic Face and Gesture Recognition, Vol.100, 2004,
pp.499-504.
[18] F. Bourel, C. C. Chibelushi, and A. A. Low, "Robust Facial Feature Tracking," in
British Machine Vision Conference. Bristol, 2000, pp.232-241.
[19] K. Morik, P. Brockhausen, and T. Joachims, "Combining statistical learning with a
knowledgebased
approach -- A case study in intensive care monitoring," in 16th International Conference on
Machine Learning (ICML-99). San Francisco, CA, USA: Morgan Kaufmann, 1999, pp.268-
277.
[20] S. Singh and N. Papanikolopoulos, "Vision-based detection of driver fatigue,"
Department of Computer Science, University of Minnesota, Technical report 1997.
[21] D. N. Metaxas, S. Venkataraman, and C. Vogler, "Image-Based Stress Recognition
Using a Model-
Based Dynamic Face Tracking System," International Conference on Computational
Science, pp.813-821,
2004.
[22] M. M. Rahman, R. Hartley, and S. Ishikawa, "A Passive And Multimodal Biometric
System for
Personal Identification," in International Conference on Visualization, Imaging and Image
Processing.
Spain, 2005, pp.89-92.
[23] R. Brunelli and D. Falavigna, "Person identification using multiple cues," IEEE
Transactions on Pattern Analysis and Machine Intelligence, Vol.17, pp.955-966, 1995.
[24] M. Viswanathan, H. S. M. Beigi, A. Tritschler, and F. Maali, "Information access using
speech, speaker and face recognition," in IEEE International Conference on Multimedia and
Expo, Vol.1, 2000, pp.
493--496.
[25] A. K. Jain, K. Nandakumar, X. Lu, and U. Park, "Integrating Faces, Fingerprints, and
Soft Biometric
Traits for User Recognition," Proceedings of Biometric Authentication Workshop, in
conjunction with ECCV2004,LNCS 3087, pp.259-269, 2004.
Page 60
51
[26] P. Melin and O. Castillo, "Human Recognition using Face, Fingerprint and Voice," in
Hybrid Intelligent Systems for Pattern Recognition Using Soft Computing, Vol.172, Studies
in Fuzziness and Soft Computing: Springer Berlin / Heidelberg, 2005, pp.241-256.
[27] K. Chang, K. W. Bowyer, S. Sarkar, and B. Victor, "Comparison and Combination of
Ear and Face
Images in Appearance-Based Biometrics," IEEE Transactions on Pattern Analysis and
Machine
Intelligence, Vol.25, pp.1160-1165, 2003.
[28] R. Chellappa, A. Roy-Chowdhury, and S. Zhou, "Human Identification Using Gait and
Face," in The
Electrical Engineering Handbook, 3rd ed: CRC Press, 2004.
[29] S. Ben-Yacoub, J. Luttin, K. Jonsson, J. Matas, and J. Kittler, "Audio-visual person
verification," in IEEE Computer Society Conference on Computer Vision and Pattern
Recognition, Vol.1. Fort Collins, CO,
USA, 1999, pp.580-585.
[30] X. Zhou and B. Bhanu, "Feature fusion of side face and gait for video-based human
identification,"
Pattern Recognition, Vol.41, pp.778-795, 2008.
[31] D. Bouchaffra and A. Amira, "Structural hidden Markov models for biometrics: Fusion
of face and fingerprint," Pattern Recognition, Vol.41, pp.852-867, 2008.
[32] H. Vajaria, T. Islam, P. Mohanty, S. Sarkar, R. Sankar, and R. Kasturi, "Evaluation and
analysis of a face and voice outdoor multi-biometric system," Pattern Recognition Letters,
Vol.28, pp.1572-1580, 2007.
[33] Y.-F. Yao, X.-Y. Jing, and H.-S. Wong, "Face and palmprint feature level fusion for
single sample
biometrics recognition," Neurocomputing, Vol.70, pp. 1582-1586, 2007.
[34] J. Zhou, G. Su, C. Jiang, Y. Deng, and C. Li, "A face and fingerprint identity
authentication system based on multi-route detection," Neurocomputing, Vol.70, pp.922-931,
2007.
[35] C. Nastar and M. Mitschke, "Real time face recognition using feature combination," in
Third
Page 61
52
IEEE International Conference on Automatic Face and Gesture Recognition. Nara, Japan,
1998, pp. 312-317.
[36] S. Gong, S. J. McKenna, and A. Psarrou., Dynamic Vision: From Images to Face
Recognition: Imperial College Press (World Scientific Publishing Company), 2000.
[37] T. Jebara, "3D Pose Estimation and Normalization for Face Recognition," Center for
Intelligent
Machines, McGill University, Undergraduate Thesis May, 1996.
[38] P. J. Phillips, H. Wechsler, J.Huang, and P. J. Rauss, "The FERET database and
evaluation procedure for face-recognition algorithm," Image and Vision Computing, Vol.16,
pp.295-306, 1998.
[39] D. Blackburn, J. Bone, and P. J. Phillips, "Face recognition vendor test 2000," Defense
Advanced
Research Projects Agency, Arlington, VA, Technical report A269514, February 16, 2001.
[40] P. J. Phillips, P. Grother, R. J. Micheals, D. M. Blackburn, E. Tabassi, and J. M. Bone,
"Face
Recognition Vendor Test (FRVT 2002)," National Institute of Standards and Technology,
Evaluation
report IR 6965, March, 2003.
[41] K. Messer, J. Kittler, M. Sadeghi, M. Hamouz, A. Kostin, F. Cardinaux, S. Marcel, S.
Bengio, C.
Sanderson, J. Czyz, L. Vandendorpe, C. McCool, S. Lowther, S. Sridharan, V. Chandran, R.
P. Palacios, E.
Vidal, L. Bai, L. Shen, Y. Wang, Y.-H. Chiang, H.-C. Liu, Y.-P. Hung, A. Heinrichs, M.
Müller, A. Tewes,
C. v. d. Malsburg, R. P. Würtz, Z. Wang, F. Xue, Y. Ma, Q. Yang, C. Fang, X. Ding, S.
Lucey, R. Goss, H.
Schneiderman, N. Poh, and Y. Rodriguez, "Face Authentication Test on the BANCA
Database," in
17th International Conference on Pattern Recognition, Vol.4. Cambridge, UK, 2004,
pp.523-532.
Page 62
53
[42] X. Q. Ding and C. Fang, "Discussions on some problems in face recognition," in
Advances In
Biometric Person Authentication, Proceedings, Vol. 3338, Lecture Notes In Computer
Science: Springer
Berlin / Heidelberg, 2004, pp.47-56.
[43] J. Yang, X. Chen, and W. Kunz, "A PDA-based face recognition system," in
Proceedings of sixth IEEE Workshop on Applications of Computer Vision. Orlando, Florida,
2002, pp.19-23.
[44] W. Zhao, R. Chellappa, P. Phillips, and A. Rosenfeld, "Face Recognition: A Literature
Survey," ACM
Computing Surveys, Vol.35, pp.399-458, 2003.
[45] A. F. Abate, M. Nappi, D. Riccio, and G. Sabatino, "2D and 3D face recognition: A
survey," Pattern
Recognition Letters, Vol.28, pp.1885-1906, 2007.
[46] R. Brunelli and T. Poggio, "Face recognition: features versus templates," IEEE
Transactions on Pattern Analysis and Machine Intelligence, Vol.15, pp.1042-
1052, 1993.
[47] M. A. Grudin, "On internal representations in face recognition systems," Pattern
Recognition, Vol.33,
pp.1161-1177, 2000.
[48] B. Heisele, P. Ho, J. Wu, and T. Poggio, "Face recognition: component-based versus
global approaches," Computer Vision and Image Understanding, Vol.91, pp.6-21, 2003.
[49] T. Kanade, "Picture Processing System by Computer Complex and Recognition of
Human Faces," Kyoto University, Japan, PhD. Thesis 1973.
[50] A. Yuille, D. Cohen, and P. Hallinan, "Feature extraction from faces using deformable
templates,"
in IEEE Computer Society Conference on Computer Vision and Templates. San Diego, CA,
USA, 1989,
pp.104-109.
[51] N. Roeder and X. Li, "Experiments in analyzing the accuracy of facial feature
detection," Vision Interface '95, pp.8-16, 1995.
Page 63
54
[52] C. Colombo, A. D. Bimbo, and S. D. Magistris, "Human-computer interaction based on
eye movement
tracking," Computer Architectures for Machine Perception, pp.258-263, 1995.
[53] M. Nixon, "Eye spacing measurement for facial recognition," in SPIE Proceedings,
1985, pp.279-285.
[54] D. Reisfeld, "Generalized symmetry transforms: attentional mechanisms and face
recognition," Tel-
Aviv University, PhD. Thesis, technical report 1994.
[55] H. P. Graf, T. Chen, E. Petajan, and E. Cosatto, "Locating faces and facial parts," in
International
Workshop on Automatic Face- and Gesture-Recognition, 1995, pp.41-46.
[56] I. Craw, D. Tock, and A. Bennett, "Finding face features," in Second European
Conference on
Computer Vision, 1992, pp.92-96.
[57] I. J. Cox, J. Ghosn, and P. N. Yianilos, "Featurebased face recognition using mixture-
distance," in
Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, 1996,
pp.209-216.
[58] S. Lawrence, C. L. Giles, A. C. Tsoi, and A. D. Back, "Face Recognition: A
Convolutional Neural Network Approach," IEEE Transactions on Neural Networks, Special
Issue on Neural Networks and Pattern Recognition, pp.1-24, 1997.
[59] L. Wiskott, J.-M. Fellous, N. Krüger, and C. von der Malsburg, "Face Recognition by
Elastic Bunch
Graph Matching," IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol.19,
pp.775-
779, 1997.
[60] M. Lades, J. C. Vorbrüggen, J. Buhmann, J. Lange, C. v. d. Malsburg, R. P. Würtz, and
W. Konen,
"Distortion invariant object recognition in the dynamic link architecture," IEEE Trans.
Computers, Vol.42, pp.300-311, 1993.
Page 64
55
[61] L. Wiskott, J. M. Fellous, N. Krüger, and C. von der Malsburg, "Face Recognition by
Elastic Bunch
Graph Matching," in Intelligent Biometric Techniques in Fingerprint and Face Recognition,
L. C. Jain, U.
Halici, I. Hayashi, S. B. Lee, and Jae-Ho, Eds.: CRC Press, 1999, pp.355-396.
[62] P. J. Phillips, P. Rauss, and S. Der, "FERET (FacE REcognition Technology)
Recognition Algorithm
Development and Test Report," U.S. Army Research Laboratory ARL-TR-995, 1996.
[63] P. J. Phillips, H. Moon, S. A. Rizvi, and P. J. Rauss, "The FERET Evaluation
Methodology for Facerecognition Algorithms," in Proceedings, IEEE Conference on
Computer Vision and Pattern Recognition, 1997, pp.137-143.
[64] G. Sukthankar, "Face recognition: a critical look at biologically-inspired approaches,"
Carnegie Mellon
University, Pittsburgh, PA, Technical Report: CMURITR-00-04 2000.
[65] P. Campadelli and R. Lanzarotti, "A Face Recognition System Based on Local Feature
Characterization," in Advanced Studies in Biometrics, Vol.3161, LectureNotes in Computer
Science, M. Tistarelli, J. Bigun, and E. Grosso, Eds. Berlin: Springer, 2005, pp.147-152.
[66] H. Shin, S. D. Kim, and H. C. Choi, "Generalized elastic graph matching for face
recognition," Pattern
Recognition Letters, Vol.28, pp.1077–1082, 2007.
[67] A. Albiol, D. Monzo, A. Martin, J. Sastre, and A. Albiol, "Face recognition using HOG–
EBGM,"
Pattern Recognition Letters, Vol.29, pp.1537-1543,2008.
[68] L. D. Harmon, M. K. Khan, R. LAsch, and P. F. Raming, "Machine Identification of
human faces,"
Pattern Recognition, Vol.13, pp.97-110, 1981.
[69] L. D. Harmon, S. C. Kuo, P. F. Raming, and U. Raudkivi, "Identification of human face
profiles by
computers," Pattern Recognition, Vol.10, pp.301-312, 1978.
[70] G. J. Kaufman and K. J. Breeding, "Automatic recognition of human faces from profile
silhouettes,"
Page 65
56
IEEE Transactions On Systems Man And Cybernetics, SMC, Vol.6, pp.113-121, 1976.
[71] Z. Liposcak and S. Loncaric, "A scale-space approach to face recognition from
profiles," in Proceedings of the 8th International Conference on Computer Analysis of
Images and Patterns, Vol. 1689, Lecture Notes In Computer Science. London, UK: Springer-
Verlag, 1999, pp.243-250.
[72] Z. Liposcak and S. Loncaric, "Face recognition from profiles using morphological
signature transform," in Proceedings of the 21st Int'l Conference Information Technology
Interfaces. Pula, Croatia, 1999, pp.93-98.
[73] R. Brunelli and T. Poggio, "Face Recognition Through Geometrical Features," in
Proceedings of the Second European Conference on Computer Vision, Vol.588, Lecture
Notes In Computer Science, G. Sandini, Ed. London, UK: Springer-Verlag, 1992, pp.782-
800.
[74] R. Cendrillon and B. C. Lowell, "Real-Time Face Recognition using Eigenfaces," in
Proceedings of the SPIE International Conference on Visual Communications and Image
Processing, Vol.4067, 2000, pp.269-276.
[75] R. J. Baron, "Mechanisms of Human Facial Recognition," International Journal of
Man-Machine Studies, Vol.15, pp.137-178, 1981.
[76] R.-J. J. Huang, "Detection Strategies for face recognition using learning and evolution,"
George
Mason University, Fairfax, Virginia, Ph. D. Dissertation 1998.
[77] L. Sirovich and M. Kirby, "Low-dimensional Procedure for the Characterization of
Human Faces," Journal of the Optical Society of America A: Optics, Image Science, and
Vision, Vol.4, pp.519-524, 1987.
[78] A. K. Jain and R. C. Dubes, Algorithms for Clustering Data. New Jersey: Prentice-Hall,
1988.
[79] K. Fukunaga, Introduction to Statistical Pattern Recognition, second ed. Boston, MA:
Academic
Press, 1990.
[80] M. Turk and A. Pentland, "Face Recognition Using Eigenfaces," in Proceedings of the
IEEE Conference on Computer Vision and Pattern Recognition, 1991, pp.586-591.
Page 66
57
[81] M. Turk and A. Pentland, "Eigenfaces For Recognition," Journal Of Cognitive
Neuroscience, Vol.3, pp.71-86,1991.
[82] A. Pentland, B. Moghaddam, and T. Starner, "Viewbased and modular eigenspaces for
face recognition," in IEEE Conference on Computer Vision and Pattern Recognition, 1994,
pp.84-90.
[83] P. N. Belhumeur, J. P. Hespanha, and D. J. Kriegman, "Eigenfaces vs. Fisherfaces:
Recognition using class specific linear projection," IEEE Transactions on Pattern Analysis
and Machine Intelligence, Vol.19,
pp.711-720, 1997.
[84] Y. Moses, Y. Adini, and S. Ullman, "Face recognition: the problem of compensating for
changes in
illumination direction," in European Conf. Computer Vision, 1994, pp.286-296.
[85] R. A. Fisher, "The use of multiple measures in taxonomic problems," Annals of
Eugenics, Vol.7, pp. 179-188, 1936.