International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169

Volume: 5 Issue: 5 1108 1111

_______________________________________________________________________________________________

Face Recognisition Using PCA Method

Rajaneesh Ganiger1,Nanda Hunagund2, Shruti Surangi3, Suman Patil4, Vidyavathi M5
1,2,3,4
Student,5Professor
Department of Electronics & communication Engineering
Smt. Kamala & Sri Venkappa M Agadi College of Engineering and Technology

Abstractthis paper essentially addresses the working of face acknowledgment framework by utilizing Principal Component Analysis
(PCA). PCA is a factual approach utilized for diminishing the quantity of factors in face acknowledgment. In PCA, each picture in the
preparation set is spoken to as a direct blend of weighted eigenvectors called eigenfaces. These eigenvectors are acquired from covariance
network of a preparation picture set. The weights are discovered subsequent to choosing an arrangement of most applicable Eigenfaces.
Acknowledgment is performed by anticipating a test picture onto the subspace crossed by the eigenfaces and afterward characterization is
finished by measuring least Euclidean separation. Various tests were done to assess the execution of the face acknowledgment framework.

IndexTermsPCA,Eigenvalue,Eigenvector,Covariance,Euclideandistance,Eigenface.

__________________________________________________*****_________________________________________________

I. INTRODUCTION arrangement of fundamental attributes, eigenfaces, which

In the course of the most recent ten years or somewhere in are the principle parts of the underlying arrangement of
the vicinity, confront acknowledgment has turned into a learning pictures (preparing set). Acknowledgment is
famous zone of research in PC vision and a standout finished by anticipating another picture in the eigenface
amongst the best uses of picture examination and subspace, after which the individual is arranged by
comprehension. In view of the way of the issue, not just contrasting its position in eigenface space and the position
software engineering specialists are occupied with it, yet of known people [3]. The upside of this approach over other
neuroscientists and therapists moreover. It is the general face acknowledgment frameworks is in its effortlessness,
assessment that advances in PC vision research will give speed and lack of care to little or progressive changes on the
helpful bits of knowledge to neuroscientists and analysts face. The issue is constrained to documents that can be
into how human mind functions, and the other way around utilized to perceive the confront. To be specific, the pictures
[1].The objective is to execute the framework (demonstrate) must be vertical frontal perspectives of human
for a specific face and recognize it from an expansive countenances. The entire acknowledgment prepare includes
number of put away faces with some continuous varieties two steps:
also. It gives us productive approach to discover the lower
The Initialization process involves the following operations:
dimensional space. Encourage this calculation can be
i. Acquire the initial set of face images called as
reached out to perceive the sexual orientation of a man or to
training set.
translate the outward appearance of a man.
ii. Calculate the Eigenfaces from the training set,
Acknowledgment could be advancement and exactness. This
keeping only the highest eigenvalues. These M
approach is favored because of its effortlessness, speed and
images define the face space. As new faces are
learning capacity [2]. Completed under broadly shifting
experienced, the eigenfaces can be updated or
conditions like frontal view, a 45 view, scaled frontal view,
recalculated.
subjects with scenes and so forth are attempted, while the
iii. Calculate distribution in this M-dimensional space
preparation informational collection covers restricted
for each known person by projecting his or her
perspectives. The calculation models the continuous
face images onto this face-space.distribution in this
differing lighting conditions as well. In any case, this is out
M-dimensional space for each known person by
of extent of the present usage. The point of this examination
paper is to contemplate and build up an effective MATLAB
These operations can be performed every once in a while at
program for face acknowledgment utilizing important
whatever point there is a free abundance operational limit.
segment investigation and to perform test for program
This information can be stored which can be utilized as a
II. FACE RECOGNITION PROCESS part of the further strides dispensing with the overhead of
One of the least difficult and best PCA approaches utilized re-instating, diminishing execution time consequently
in face acknowledgment frameworks is the supposed expanding the execution of the whole framework [4]
eigenface approach. This approach changes faces into a little
1108
IJRITCC | May 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 5 1108 1111
_______________________________________________________________________________________________
Having initialized the system, the next process involves the as eigenfaces. We normalize the eigenvectors so that
steps: they are orthonormal, The eigenvectors are ordered so
i. Calculate a set of weights based on the input image that 2 i > 2 1+1 .The Aa s are equal to the variance of the
and the M eigenfaces by projecting the input projection of the training set onto the itheigenvector.
image onto each of the Eigenfaces. Thus, the low-order eigenvectors encode the larger
ii. Determine if the image is a face at all (known or variations in the training set (low order refers to the index
unknown) by checking to see if the image is of the eigenvectors and eigenvalues). The higher-order
sufficiently close to a free space. eigenvectors encode smaller variations in the training set.
iii. If it is a face, then classify the weight pattern as Because these features encode smaller variations, it is
either a known person or as unknown. commonly assumed that they represent noise in the
iv. Update the eigenfaces or weights as either a training set. Because of this assumption and empirical
known or unknown, if the same unknown person results, higher-order eigenvectors are excluded from the
face is seen several times then calculate the representation. Faces are represented by their
characteristic weightpattern and incorporate into projection onto a subset of M < N 1 eigenvectors,
known faces. The last step is not usually a which we will call face space(see figure 1). Thus, a
requirement of every system and hence the steps facial image is represented as a point in an M -
are left optional and can be Implemented as when dimensional face space. The dimension M is
the there is a requirement. arepresentation is (N-1) dimensional because the
requirement that E t, = 0 removes one degree of
III. EIGENFACE ALGORITHM freedom.A face is represented by its projection onto a
subset of M eigenvectors. A set of facial images
becomes a set of points '0' in face space. The point
marked by 'X' is a probe and is identified as the person
in the gallery image nearest 'X'.Design decision that is
discussed in the paper. A gallery of K facial images is
represented as K po ints {g 1 , g i c in face sp ace.
A probe is identified by first projecting it
into face space and then comparing the projection to
all gallery images. We denote a probe by p t . A probe is
compared to gallery images by a similarity measure.
The similarity between probe p i and gallery image g k
is denoted by s ; (k). Two possible similarity measures
are the Euclidean and L i distances between p i and
g k .The identity of a p robe is determined to be the
gallery face, g k , that minimizes the similarity
measure between p i and the g k s. In this paper we
assume that there is one image per person in the
gallery, and g k * uniquely references the identity of
the person. This recognition technique is called a
nearest-neighbour classifier-a probe is identified as the
person in the gallery image nearest the probe in face
In a PCA-based face-recognition algorithm, the
space. A probe is identified by first projecting it
input is a training set, t 1 , t N of N fac ia l i ma g es
into face space and then comparing the projection to
s uc h t ha t t he e n se mb le me a n o f t h e tr a i ni n g s e t
all gallery images. We denote a probe by p t . A probe is
is zero ( t i = 0 ) . In computing the PCA
compared to gallery images by a similarity measure.
representation, each image is interpreted as a point in
The similarity between probe p i and gallery image g k
1R"' where each image is n by m pixels. PCA finds the
is denoted by s ; (k). Two possible similarity measures
optimal linear least-squares representation in(N-
are the Euclidean and L i distances between p i and
1)dimensional space, with the representation
g k .The identity of a probe is determined to be the
preserving variance. The PCA representation is
gallery face, g k , that minimizes the similarity
characterized by a set of N 1 eigenvectors (e 1 ,
measure between p i and the g k s. In this paper we
eN_i)and eigenvalues (2 1, ...). In the face-
assume that there is one image per person in the
recognition literature, the eigenvectors can bereferred to
gallery, and g k * uniquely references the identity of
1109
IJRITCC | May 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 5 1108 1111
_______________________________________________________________________________________________
the person. This recognition technique is called a data type for this package, but handles are predefined for the
nearest-neighbour classifier-a probe is identified as the GUI in the GUI_OpeningFcn and hence there data type
person in the gallery image nearest the probe in face cannot be modified. Instead of passing handles in those
space. functions we can save the image as a jpg on the disk and
Our face-recognition system consists of three then later on use the main function to load that picture. This
modules and each module is composed of a sequence of is not as convenient as passing handles as more data will be
steps (see figure 2). The first module normalizes the required to be saved to the disk i.e. coordinates, names,
input image. The goal of the normalization module is to counters. Due to all these factors, a Flash application seems
transform the facial image into a standard format that to be the best choice, because the data is still required to be
removes or attenuates variations that can affect saved on the disk by MATLAB so it can be accessed later
recognition performance. This module consists of four on. Flash also provides an opportunity to make the
steps; figure 3 shows the results of processing for application look more user friendly and attractive.s
some of the steps in the normalization module, The first
step low-pass filters or compresses the original image.
Images are filtered to remove high-frequency noise.
An image is compressed to save storage space and
reduce transmission time. The second step places the face
in a standard geometric position by rotating, scaling, and
translating the center of eyes to standard locations. The
goal of this step is to remove variations in size,
orientation, and location of the face in an image. The
third step masks background pixels, hair, and
clothes.

Figure 11.1 Selection of images

IV. GUI IMPLEMENTATIONS

Two approaches to implement the GUI were considered.
First one was to implement it in MATLAB. Although
developing the GUI in MATLAB should be faster, its not
user attractive enough. Another major issue are the GUI
handles in MATLAB. Normally the main GUI structure is
written first and then where necessary the handles are passed
to the function who wants to plot. The problem arose when Figure (a)
the handles were required by localframecallback function.
This function is part of the video object and gets a new
frame from the video acquisition device. It only expects two
arguments and it cant be redefined to contain three
arguments to include handles as its the object property
which is fixed by MATLAB. A possible solution was to
save the handles in a MAT file and then load that file inside
the function. This didnt succeed as the link between the
handles and the GUI objects was not preserved when the
data type was loaded. Another possible solution was to use Figure (b)
persistent data type for handles which is a global variable

1110
IJRITCC | May 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 5 1108 1111
_______________________________________________________________________________________________
demonstrated that this strategy gave great grouping of
appearances however it has restrictions over the varieties in
size of picture. The eigenface approach consequently gives a
down to earth arrangement that is very much fitted to the
issue of face acknowledgment. It is quick, moderately
straightforward what's more, has been appeared to function
Figure (c) admirably in obliged condition.

VI. FUTURE PLAN

In this thesis paper, we worked with some still pictures but
we will try to develop a system using video camera that will
work with real time face recognition. We want to overcome
the problem of different size face image recognition. We
will compare the performance analysis of PCA based
method with all others existing face recognition methods.
Figure (d)
REFERENCES
[1] http://www.face-rec.org
[2] Shemi P M, Ali M A, A Principal Component Analysis
Method for Recognition of Human Faces: Eigenfaces
Approach, International Journal of Electronics
Communication and Computer Technology
(IJECCT),Volume 2 Issue 3 (May 2012).
Figure(e) [3] M. Turk, A. Pentland: Face Recognition using Eigenfaces,
Figure a, b, c, d, e Steps in face detection process. Conference on Computer Vision and Pattern Recognition,
3 6 June 1991, Maui, HI , USA, pp. 586 591.
[4] Prof. Y. VijayaLata , Chandra Kiran
BharadwajTungathurthi , H. Ram Mohan Rao , Dr. A.
Govardhan , Dr. L. P. Reddy, Facial Recognition using
Eigenfaces by PCA, Department of Computer Science and
Engineering, GokarajuRangaraju Institute of Engg&Tech,
Jawaharlal Nehru Tech. University.
[5] Parvinder S. Sandhu, Iqbaldeep Kaur, Amit Verma, Samriti
Jindal, Inderpreet Kaur, ShilpiKumari, Face Recognition
Using Eigen face Coefficients and Principal Component
Analysis, International Journal on Electrical and
Electronics Engineering 3:8 2009.

Figure: Output of face detection

V. CONCLUSION
In this postulation we actualized the face acknowledgment
framework utilizing Principal Component Analysis and
Eigenface approach. The framework effectively perceived
the human confronts and worked better in various states of
face introduction. In this exploration, Principal part
examination way to deal with the face acknowledgment
issue was considered and a face acknowledgment
framework in view of the eigenfaces approach was
proposed. The calculation created in a summed one up
which functions admirably with a pictures. The tests
directed on Bitmap pictures, PNG pictures and JPEG
pictures of different subjects in various postures

1111
IJRITCC | May 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________