Face Recognition Using Principal Component Analysis
Ramandeep Kaur
Computer science and engineering
CT institute of Technology and Research (CTITR),
Maqsudan, Jalandhar, India
ramandeep_wahla@yahoo.com
Abstract-The strategy of face recognition involves the
examination of facial features in a picture, recognizing those
features and matching them to 1 of the many faces in the database.
There are lots of algorithms effective at performing face
recognition, such as for instance: Principal Component Analysis,
Discrete Cosine Transform, 3D acceptance methods, Gabor
Wavelets method etc. This work has centered on Principal
Component Analysis (PCA) method for face recognition in an
efficient manner. There are numerous issues to take into account
whenever choosing a face recognition method. The main element is:
Accuracy, Time limitations, Process speed and Availiability. With
one of these in minds PCA way of face recognition is selected
because it is really a simplest and easiest approach to implement,
extremely fast computation time. PCA (Principal Component
Analysis) is an activity that extracts the absolute most relevant
information within a face and then tries to construct a
computational model that best describes it.
Index terms - Face Recognition, PCA, Eigen Vector and feature
extraction.
I. INTRODUCTION
Now a day face recognition continuous in demand in image
processing system because of its recuriements such as
authentication, access control, survellieance systems. Faces can
be represented on the basis of two categories such as appearance
and features. Face recognition procedure divided into two steps:
A. Face detection:
To detect whether an image represent a face i.e. the general
structure of face or not.general structure include
forehead,eyes,nose,lips,chin etc.
B. Face recognition:
Face detection is easier than face recognition, face detection is
used to detect a face from an image but face recognition is used
to find out whose face is this.in order to recognize a face no.
there are no. of techniques such as PCA (principal component
analysis), LBP (local binary patterns), ICA (independent
component analysis) LDA (linear discriminant analysis), and so
on. These techniques are used for face recognition. Some kind of
978-1-4799-8047-5/15/$31.00 2015
c IEEE
Er. Himanshi
Computer science and engineering
CT Institute of Technology and Research (CTITR),
Maqsudan, Jalandhar, India
variations occurred that creates difficulties in face detection and
the variations are illumination change, pose variation, RST
variation, Cluttering, Occlusion. Face recognition system can
help in many ways. It is used to check criminal records, pattern
recognition, enhancement of security using cameras etc.
Zahid Riaz et al. (2005), [1] have proposed approach of
classifying the humans on the basis of their compressed face
images. Classification technique utilizes PCA in certain different
way. Only first principal component is employed as feature
vector out of 92 components (since image size is 112x92),
causing an improved result of 87.39%. D.A. Meedeniya et al.
(2007), [2] have proposed a new method for face recognition.
This process depends on an alternative of utilizes Principal
Component Analysis (PCA) technique. The algorithm extracts
the Eigen prices and Eigen vectors from the images. It functions
the financial measurement single price decomposition to acquire
a unitary matrix, which will be use for recognition. The pictures
are acknowledged bottom on the minimum distance. This
algorithm can achieve 93.7% and higher performance.
Janarthany Nagendrarajah et al. (2010), [3] have proposed This
model proposes an approach to handle recognition where in fact
the facial expression in working out image and in the testing
image diverge and merely a single sample image per class can
be acquired to the machine. This technique has been used on a
few pictures to have the ability to remove a few Eigen-images
known as Eigen encounters and loads with this particular
illustration are applied for recognition. For the classification job,
range full Euclidean Range has been applied to acquire the exact
distance with the fat vectors attached with the education images.
When tried with ten topics and six standard words the full
acceptance cost was 89%, for skilled encounters. Patrik
Kamencay et al. (2012), [4] has centered on preprocessing face
images using segmentation algorithm and SIFT (Scale Invariant
Feature Transform) descriptor the paper presents a proposed
methodology for face recognition. Karande Kailash J. et al.
(2012), [5] the situation of experience acceptance applying
Laplacian pyramids with various orientations and separate parts
is resolved in that paper. The edginess like data is received by
585
 utilizing Concentrated Laplacian of Gaussian (OLOG) strategies
with four various orientations (00, 450, 900, and 1350) then
preprocessing is performed by utilizing Concept Part evaluation
(PCA) before acquiring the Separate Components. The separate
parts received by ICA methods are utilized as function vectors
for classification. The Euclidean range (L2) classifier is useful
for screening of photographs. Lingling Peng et al. (2012), [6]
have proposed a Theory Vectors Subspace (PVS) for experience
recognition. Firstly, use PCA to acquire each aspect vector;
therefore we attain a subspace which ends theory vectors of
every aspect. In order to evaluate the performance make a
comparison of PCA, KPCA on the ORL and AR databases.
Rammohan Mallipeddi et al. (2012) ,[7] have proposed, the NP-
hard issue of obtaining the most effective part of the removed
PCA characteristics for experience acceptance is resolved
utilizing the differential development (DE) algorithm and is
known as FS-DE. Also provide a collection centered method for
experience acceptance (En-FR), wherever various subsets of
PCA characteristics are purchased by maximizing the exact
distance between a part of lessons of working out information in
place of full lessons. Shiji S. k et al. (2013), [8] have proposed
face recognition is completed by Principal Component Analysis
(PCA). Face images are projected onto an experience space that
encodes best variation among known face images. The face area
space is defined by Eigen face which can be eigenvectors of the
group of faces that might not match general facial features such
as for instance eyes, nose, and lips. The Eigen face approach
uses the PCA for recognition of the images. Computers that
recognize and detect faces might be placed on a wide selection
of practical applications including security systems, criminal
identification, identity verification etc. Zhiyong Zeng et al.
(2013), [9] have proposed a new method for face description and
recognition using shearlets transform and principle component
analysis. Finally, face feature is extracted by principle
component analysis. Ban Jozef et al. (2013) , [10] have give a
comparative study of several conventional face recognition
methods (PCA a.k.a. Eigen faces, RBF) and novel kernel
methods (KPCA, GDA and SVM) which are suitable to work
effectively beneath the conditions. This paper centered on
occlusions of eyes and eyebrows as they are probably the most
significant options that come with a face. Face recognition rates
accomplished by machine learning methods with accuracy
accomplished by human perception only are compared. Tomesh
Verma et al. (2013) , [11] have proposed face recognition is
conducted using Principal Component Analysis followed closely
by Linear Discriminant Analysis based dimension reduction
techniques. Sequencing of the paper is pre processing, PCA is
used for dimension reduced total of training database set , LDA
is used for extraction of features for class separabilityand finally
testing by nearest mean classification techniques. Method is
tested over ORL face database. It is located that recognition rate
with this database is 96.35% and hence showing efficiency of
586 2015 IEEE International Advance Computing Conference (IACC)
the proposed method than previously adopted types of face
recognition systems.
II. TECHNIQUES OF FACE RECOGNITION
There are following techniques that are used for face
recognition:-
A. PCA (Principal Component Analysis)
PCA is a standard method used for recognition of statistical
design in order to reduce dimensionality and used for feature
extraction. It is used to preserve the important information of the
pattern and is used to remove redundant information. Given a
database contains no. of images; the task of a face recognition
system is to show that the picture on which testing is performed
belongs to a person in the respositary. A face contains certain set
of features and these characteristic features are called principal
components or Eigen faces. These features can be extarcted
from the original image with the help of principal component
analysis. Appearance of a face image of size M X M can be
considering as M2 1-dimensional vector in order to reduce
dimensionality.
B. LDA (Linear Discriminant Analysis)
LDA is also called fisherface method. It tries to improve the
proportion of determinat of between classes/ determinant of
within school spread matrix of expected products. Following
example show the good class separation and the bad class
separation. With the help of fisher discriminant, grouping the
images belongs to the same classes, separating the images
belongs to different classes. Convert the dimesions of face
images from N2 dimension space to C dimension space, C is the
number of classes of images while projecting the face images on
the fisher space. For example two set of points are placed in two
dimesional space onto a singe line, depending upon the direction
points are either mixed or separated. Fisher discriminant is used
to find out a line which best separate points in order to find out
the test images. Perform the comparison of test images with the
database images and find out which database image is close to
the test image for recognition purpose.
Between
Between
Within Within
(a) (b)
Fig. 1. (a) Good class separation (b) Bad class separation
C. ICA (Independent Component Analysis)
PCA does not support higher order dependencies, it is based
on the second order satistics of the image set. ICA is the
generalization of the PCA, in addition to the second order
statistics ICA separates the higher order moments. Architecture I
take the image as random variables and provide the outcomes as
pixels. Architectue II produces the independent coding variables
(coefficient). By the product of individual probabalties, the
probability of combination of features can be obtained that
provide the factorial facial code. ICA representation is higer
order rank to PCA representation that is used for face
recognition under different variations.
D. Active Appearance Model
Apply this model after collecting several face images that
have different alligment features. This model is used to provide
accurate alligment of the features as well as for the pose
correction.
E. LBP (Local Binary Pattern)
Local binary patterns provide the textute as well as the shape
of the digital image; this is done by dividing a digital image into
several small regions in order to extract feature. These features
contains the binary patterns that provide the surrounding of
pixels in the region. Image is represented by representing
obtained features from the region into a single feature histogram.
Ojala et al introduce the original LBP operator that represent
texture of an image by labeling thresholding to pixels of an
image by 3 x 3 neighbours of each pixel with the centre value
and provides results as binary number. Thereafter labels
histogram are used to define texture of an image. The following
figure show a digital face image is divided into the 64 small
regions:
when selecting an experience recognition method. The keys ones
were: Accuracy, Time limitations, Process speed and
Availability. With one of these in mind the PCA approach to
face recognition was selected because it is
• Easiest solution to implement.
• Extremely fast computation time.
• Accurate.
There are many disadvantages with the PCA method are: PCA
is
• Translation variant
• Scale variant
• Background variant
• Lighting variant
A. Theory of PCA
Principal Component Analysis is an activity that extracts the
absolute most relevant information within a face and then
attempt to construct a computational product that best describes
it. Given a database contains no. of images; the task of a face
recognition system is to show the check picture goes to an
individual in the repository. A face contains certain set of
features and these characteristic features are called principal
components. These features can be extarcted from the original
image with the help of principal component analysis.
Flow chart for face recognition using PCA approach shows
the different steps used for the face recognition purposes.
Input image Trained image
Develop dataset

Evaluate feature
using PCA
Evaluate error

Evaluate ecludian distance

Fig. 2. Preprocessing image divided into small regions

III. RESULTS AND DISCUSSION Find minimum ecludian

distance
The method of face recognition involves the examination of
facial features in a picture, recognizing those features and
matching them to at least one of the numerous faces in the Recognize image is
database. There are lots of algorithms effective at performing minimum distance
face recognition; such as for instance: Principal Component
Analysis, Discrete Cosine Transform, 3D recognition methods,
Gabor Wavelets method etc. We've centered on Principal
Component Analysis (PCA) method for face recognition within End
an efficient manner. There have been many issues to think about

2015 IEEE International Advance Computing Conference (IACC) 587

 Through PCA it's possible to change each original picture of
working out set in to an equivalent Eigen face. An important
feature of PCA is any particular you can reconstruct any unique
image from working out set by combining the Eigen faces.The
initial face picture might be reconstructed from Eigen faces if
most of the eigenfaces (features) are added in the proper
proportion. Each Eigen face represents only certain top features
of the face area that might or may probably not be present. In the
initial picture if the function is within the unique picture to an
elevated level, the reveal of the equivalent Eigen experience in
the “sum” of the eigenfaces should be greater.
If contrary, this feature isn't (or almost not) within the initial
image, a corresponding Eigen face should contribute an inferior
(or not at all) part to the quantity of eigenfaces. Therefore, to be
able to reconstruct the first picture from the eigenfaces, you've
got to generate some sort of measured amount of all eigenfaces.
That is, the reconstructed unique picture is b0 h-equal to a
volume of all eigenfaces, with each Eigen experience having a
certain weight. That fat identifies, from what amount the
particular function (Eigen face) is the first image. If one employs
all of the eigenfaces produced from unique photographs, you can
reconstruct the first photographs from the eigenfaces exactly.
But it's probable to also use simply a main eigenfaces. A
reconstructed picture is definitely an approximation of the first
image.
3. Recognition is conducted by comparing the positioning of a
face in the eigenspace with the place of known users
Steps to find out Eigen vectors are:
Step1: The vector of all training images saved in ASCII file
ought to be called and will undoubtedly be stored in a matrix.
S = {A1, A2, A3 ……… AN} (1)
The above equation 1 shows a collection S which includes
most of the training images; here most “N” is the amount of
available training images stored in the database.
Step2: Finding mean of training images as indicated in equation
2.
μ = ேଵ σே௠ୀଵ Γ ݉
μ = Means of training set
Γ m = Vector of Training image
ܰ= Number of training images
Step 3: Deduct the mean as indicated in equation 3.
Φm = Γm − μ
Φ m =The deducted vector of training image.
Step4: Calculate the covariance matrix as indicated in equation
4.

N
1
C = ¦ Φ m
N m =1

C Is the covariance matrix which includes deducted image

vectors?
Step5: Calculate Eigen vectors and Eigen values. Eigenvectors
and corresponding Eigen values ought to be
Fig. 3. Database contain no. of images
calculated from covariance matrix C .
Step6: Obtaining Eigen faces.
After perfoming comparison of input image to the no. of
database images, resultant recognized images are shown below

Fig. 4. Test dataset

The primary idea of PCA could be described in these steps:

1. Eigenvectors or eigenfaces of the covariance matrix are
found. This is performed by training a couple of face images.
2. These eigenvectors end up being the eigenspace (subspace
made up of the Eigenvectors) by which every face is
projected on. Fig. 5. Result 1 obtained on matching the image of training Dataset with that of
test dataset.

588 2015 IEEE International Advance Computing Conference (IACC)


Fig. 6. Result 2 obtained on matching the image of training dataset with that of
test dataset.
The acccuracy rate of recognized images is shown by graph
Fig. 7. Accuracy graph
IV. CONCLUSIONS AND FUTURE WORK
This paper has focused on PCA based face recognition in an
efficient manner. There were many issues to consider when
choosing a face recognition method. The key ones were:
Accuracy, Time limitations, Process speed and Availability.
With one of these in mind the PCA way of face recognition was
selected for this project – simplest and best to implement, very
fast computation time. PCA has power to recognizing a face
with a different background is difficult. The face area detection
program is altered so that it merely not only detects a face but
extracts it to a new image, thus cutting out nearly all the
background. The rest of sections of background are insignificant
since the camera’s resolution will only keep the face in focus
because of the distance. If the light strenght changes then the
reliability of face acceptance drops. This certification process
involves recording images in front of a PC and usually the room
is lighted when a user is using a PC. The PCA based approach
has the ability to recognize faces in more efficient way. The
PCA based face recognition method has been designed and
implemented in MATLAB using image processing toolbox. The
result has shown effectiveness of the PCA based face recognize
method.
This work has not considered any improvement in the PCA
base face recognition by integrating it with certain filters,
contrast enhancement techniques or neural based training etc. so
2015 IEEE International Advance Computing Conference (IACC)
in future we will integrate the proposed technique will some
well known noise removal and image enhancement methods.
REFERENCES
[1] Zahid Riaz Arif Giggiti and Zulfqar Ali, “Classification of
compressed human face images,” by using principal component,
IEEE internation conference on ICICT, 2005, pp. 121-125.
[2] Alaa Eleyan and Hasan Demirel, “PCA and LDA based face
recognition using Feedforword neural network classifier”, In
Multimedia Content Representation, Classification and Security
Springer, 2006, pp. 199-206.
[3] D.A. Meedeniya and D.AA.C. Ratnaweera, “Enhanced face
recognition through variations of principal component analysis”,
Second international conference on industrial and information
system ICIIS, 2007, pp. 347-351.
[4] Caifeng Shan, Shaogang Gong and Peter W Mcowan, “Fcaial
Expression Recognition based on local binary pattern”, Image
and Vision Computing , 27 (2009), pp. 803–816.
[5] Janarthany Negedrara and M.U.S. Perera., “Recognition of
expression variations faces-principal component analysis based
approach for access control”, IEEE, 2010, pp. 125-129.
[6] Patrik Kamenancy, Martin Brezhan, Dominik jelsovka and
Martina Zachariasova, “Improved Face Recognition method
based segmentation algorithim using SIFT-PCA”, IEEE
international conference on TSP, 2012, pp. 758-762.
[7] Karande Kaliash J., Talbar Sanjay N., Inamdar Sandeep S., “Face
Recognition using oriented laplacian of gaussian (OLOG) &
ICA”, IEEE transaction on image processing, 2012, pp. 99-103.
[8] Lingling Peng and Qiong Kang, “Face recognition by principal
vector subspace”, IEEE, 2012, pp. 298-300.
[9] Rammohan Mallipeddi and Minho Lee, “Ensemble Based Face
Recognition using discriminant PCA features”, IEEE CEC,2012.
[10] Shiji s.K, “Biometric prediction on face images using Eigen face
approach”, IEEE conference on information and communication
technology, 2013, pp. 104-109.
[11] Zhiyongzeng and Jianqiang Hu, “Face Recognition based on
shearlets transform and principal component analysis”,
International Conference on intelligent networking and
collaborative system, 2013, pp. 697.
[12] Banjozef, Feder Matej, Omelina eubos, Oravec Milos and
Pavolovacova Jarmila, “Face Recognition under partial
occulusion and noise”, IEEE, 2013, pp. 2072-2079.
[13] Tomesh Verma and Raj Kumar Sahu, “PCA-LDA Based Face
Recognition system and result comparison by various
classification techniques”, IEEE, 2013.
[14] Eyal Arubas, “Face Detection and Recognition (Theory and
Practice)”, 6 April 2013.
589