International Journal of Engineering Trends and Technology (IJETT) - Volume4Issue4- April 2013
Biometrics Human Face Recognition
Techniques
P.RAVI TEJA#1, R.R.V.S.S.ABHISHEK *2, P.RAM MANINTH #3
Department of ECE, K L University Green Fields, Vaddeswaram, Guntur, Andhra Pradesh 522502
Abstract- Most of the technologies now being developed for the
purpose of automatic capture, measurement and identification of
distinctive physiological that could safe our identity and therefore
our property and privacy have come to be named as 'biometrics',
because they implied statistical technique to observe the
phenomena in biological. However, the followers of biometrics are
much broader than just identity verification. Biometrics plays a
essentials role in agriculture, environmental and life sciences. This
paper mainly discusses about the human face recognition with the
help of different tactics and techniques like “Eigen faces for
Recognition” and “Feature Based Recognition: Elastic Bunch
Graph Matching”. The scanning technique such as 4-D laser
scanning is still under discussion since for recognition technique it
is necessary, The real world is CCTV technology and similarity
between the human and computer recognition.
This technology is outstanding in the capturing image of faces
such as criminal investigation, terrorists identification, medical
purposes like plastic surgery .It provides almost security and
reliability compared to the other techniques. In future this
technology will be the most convenient and secure technique. This
will over through all the current security traits and become a
efficient security measure
KEYWORDS- Biometrics, CCTV technology, Eigen faces for
Recognition, Feature Based Recognition, Elastic Bunch Graph
Matching, 4-D laser scanning
INTRODUCTION
Many persons have always used individual traits for
identification. In past few years, the presence of scars,
birthmarks and other unusual features helped to minimize
mistaken identify. Even these days , we are using technology
that have been around for couple of decades , such as private
keywords like password and signatures. But passwords are
notably insecure, and signatures can be forged or duplicated.
Shop assistants, for example, often don’t usually compare the
signature on the back of a credit card or any receipt with the
sample provided by the customer. The hunt for a better way of
providing distinguished identity is discussed. The
technologies now being developed for these purposes have
come to be labeled 'biometrics', because they apply statistical
methods to biological observations and phenomena. As
computer power has grown, so too has the concept that the
automated capture, measurement and identification of
distinctive physiological characteristics
could safeguard
our identities and
ISSN: 2231-5381
Therefore our property and privacy, and could also
be used to fight crime. However, the discipline of
face identification is much broader than just identity
verification. Biometrics plays a crucial role in
agriculture, environmental and life science .Main
applications of biometrics are finger scanning, retina
scanning, size of the hand and human face
recognition.
HUMAN FACE RECOGNITION
For most of the neuroscientists and computer scientists,
face recognition is one of the most a fascinating problem
with important applications such as, surveillance, face
reconstruction. Physiological evidence indicates that the
brain consists of specialized face recognition hardware
in the form of face sensors cells in the most important
regions i.e infer temporal cortex and also in the regions
of frontal right hemisphere; impairment in these areas
leads to a syndromes popular
known as
“prosapagnasia”.
Interestingly,
“prosapagnosics”,
although was unable to known faces, retain their
functionality to visually known to non-face objects. In
computer vision, some of the most popular facial
recognition technique have been completely motivated
into biologically. Using these models, scientists can
quantify the commonly known faces; images whose
projections are close in face space are likely to be from
the same individual. We can analyze the obtained results
of these models with that of a human perceptions to
determine whether distance in face space corresponds to
the human notable of facial similarity. Although most of
the biologically inspired models are very useful for
neuro scientists, ultimately, when building a commercial
face recognition system, one should use the technique
with at most highest performance, regardless of
biological relevance. However, for certain applications,
such as witness face remodifications , in which human
face perception is mostly similarity and is relevant to the
task, models developed using human psychophysical
evidence might outperform the other techniques.
http://www.ijettjournal.org
Page 953
International Journal of Engineering Trends and Technology (IJETT) - Volume4Issue4- April 2013
NEUROPHYSIOLOGIC EVIDENCE ON FACE
RECONGITION
Few scientists theory of facial recognition said that the
outputs of ordinary edge and simple line detectors are
used as the inputs to the shape of pyramid for an higher
level of cells , to perfectly form’s complex cells that
response to every selective to a minor ranges of inputs.
At the ancestors of this pyramid could be a
“grandmother cell” that would fire only when you see
your grandmother. By contrast, the previous theories for
object recognition in the brain foreseen that objects are
encoded as changes in the firing pattern across a
population of cells, and that these cells respond to a huge
range of stimuli . Moreover these face detector cells just
simply didn’t respond to simple stimuli (e.g., textures,
bars, gratings) nor to other complex objects, including
those designed to rise emotional responses. Some of
these cells responded to minor line drawings of faces,
but couldn't response well to faces with newly
constructed or reconstructed face specification, even if
all of the specification were present. Face cells
continuously respond to faces subjected to high-pass
filters, the lower pass filters, and size scaling; colour
shifting decreased to some extent but did not eliminate
response. Consider that a total, of this body is evidence
and is fairly conclusively substantiates the existence of
face detector cells in one of the most important place i.e
primate brain. Scientists have also reported that cells
tuned to isolated facial features, such as eyes, but it is far
more ambitious to verify that these cells are specifically
tuned to eye detection, rather than to a more general
period of shapes, such as ellipses. Shifting the placement
of facial specification does somewhat affect the response
of face cells, especially on adjustment of the intraocular
distance that is the length between eyes and mouth, or
the distance of forehead hair from the forehead to the
eyes. . Cells are most commonly found in the sulcus,.
Although frontal facial detector cells are the most natural
and certain cells respond preferentially to pivot or
profiles views; interestingly, two of the computer face
recognition technologies discussed below also make use
of motion views to improve the systems performance.
Face detector cells have been reported in a huge variety
of locations, bridges all regions of the inferior temporal
gyrus and the superior temporal sulcus which can be
divide into two separate areas based on their locations
and connectivity: a superior temporal polysensory area
(STPA) and the other as inferior temporal cortex (ITC).
These two distinct populations of face detector cells may
contribute into a different circuits, one honed or
establishing identity and the other for recognizing facial
expression.
The facial recognition is most essential task for
obtaining results on primates brains ,who use facial
ISSN: 2231-5381
expressions
for different type of communication..
Thus, the existence of specialized biological
``hardware'' for face recognition is both plausible from
an evolutionary point of view and substantiated by
neurophysiological evidence. This collection of studies
neither substantiates nor disproves any of the computer
models for face recognition discussed here. Two of the
systems, Eigen face and feature-based graph matching,
have been specifically promoted as potential biological
models.
EIGENFACES ALGORITHM
``Eigen faces for Recognition'' was initial given by Turk
and Pent land’s after many researches focused on
detecting individual facial features and categorizing
different faces by the position, size, and relationship of
these features. In contrast, Turk and Pent land take an
information theoretic approach to the problem and derive
their classification features with principal component
analysis which extracts the optimal linear encoding of
the dataset. In a cluster of image vectors, the dimension
with the greatest variance is captured by the eigenvector
with the highest eigenvalue; thus when projecting a point
into the eigenvector basis and examining the value of its
coefficients, the most information about the point can be
gained from the coefficient of the top eigenvector. This
technique is also exploited by Shree and Nayar for the
general object recognition problem of classifying images
of 3-D models with variable illumination and viewpoint .
Eigenvector techniques rely mainly on being able to
classify images based mainly on the coefficients of the
top k eigenvectors.. The eigenvectors can be calculated
using an matrix, in which R is the number of images,
rather than using the original covariance matrix which
has dimensionality , where N is the number of pixels in
the image. Typically, so their technique of
premultiplying the covariance matrix is computationally
more efficient. Turk and Pent land use the following
technique to initialize their system. After acquiring an
initial training set of images, they calculate the Eigen
faces (eigenvectors) for the matrix and use those to
define a ``face space''. Each individual is represented in
this face space by averaging the coefficients extracted
from labeled exemplars of his/her face. Once the system
is initialized, images can be classified by projecting them
into the eigenvector basis, using only the top k vectors
with the highest eigenvalues. The projection is achieved
simply by taking the dot product of the image with each
eigenvector; this computation can be performed with a
three layer network in which weights are the
eigenvectors and the hidden layer's outputs are the
coefficients. The final output of the network is an image,
quite similar to the original image, that has been
reconstructed from the information stored in the
coefficients of the top k eigenvectors. Non-face images
http://www.ijettjournal.org
Page 954
International Journal of Engineering Trends and Technology (IJETT) - Volume4Issue4- April 2013
are not represented well by this basis set so the final
image will be noticeably different from the original
image, whereas a face passed through the network will
have almost identical inputs and outputs. After non face
images are discarded, the test image is assigned the label
of the closest class mean. If the Euclidean distance
between the test image and the closest mean is above a
certain threshold it is classified as an unknown face
which can be added to the dataset later. New faces can
be used to modify the Eigen faces, which unfortunately
involves recalculating all the eigenvectors and the
coefficient means. The algorithm is

The average face is defined .

The average face is then subtracted from each
and victories.

Linear combination of the face images is calculated.

Face space is defined this is compared with all faces in
data set.

It is reconstructed.
 The reconstruction error is a measure of how
typical a face this is, and is applicable for
face/non-face classification.
image
periodically added. Turk and Pent land suggest that
projecting facial images into the eigenvector basis makes
the image classes more separable because the
eigenvectors represent the dimensions of maximum
variance. Although this projection efficiently spreads the
points, it does not consider how the points are assigned
to classes. Ultimately, to project the images such that the
classes can be separated rather than the points; even if
the points are spread along an axis, one can have the
undesirable situation of a point on one end of the cluster
belonging to the same class as points on the other end of
the cluster. Averaging the coefficients of these points
will give a misleading class mean in the middle of the
cluster. In another variant of PCA, the coefficients of all
the exemplars are retained, instead of just the mean
coefficients. This variant is more successful at handling
non convex cases such as the one described in the
previous example. Methods such as Fisher's linear
discriminant analysis take class labels into account when
re-projecting the points and have been successfully
applied to the face recognition problem . Although PCA
provides a linear projection that may work for some
recognition problems, there is no reason to believe that it
is the best projection for recognition merely because it
works well for encoding.
DEVELPMENT OF FACIAL RECONGNITION
TECHNOLOGIES IN CCTV SYSTEMS
To test how this algorithm performs in changing
conditions, by varying illumination, size, and orientation
of the faces. It is found that that this system had the most
trouble with faces scaled larger or smaller than the
original dataset. To ameliorate this problem, it suggests
using a multi-resolution method in which faces are
compared to Eigen faces of varying sizes to compute the
best match. Also it is note that image background can
have a significant effect on performance, which is
minimize by multiplying input images with a 2-D
Gaussian to diminish the contribution of the background
and highlight the central facial features. This system
performs face recognition in real time and they also use
their method, along with motion cues, to segment faces
out of images by discarding squares that are classified as
non-face images. In the field of face recognition this
method is still quite popular due to its ease of
implementation. Also this algorithm can be plausibly
implemented as a biological network, and possesses
some of the characteristics of human face recognition
such as high recognition speed and robustness to
occlusion. Although this system can be fooled by short
term changes, such as variations in facial hair or
hairstyle, long term facial changes due to aging can be
handled if one assumes that new training images are
ISSN: 2231-5381
Computer software and CCTV cameras that identify
criminal faces. New ham’s Environment Department
conducted a project that has led to the production of the
automated face which is shocking to be the first face
scanning system to be ever used in public CCTV
systems . Officers in Newham began to investigate the
possibilities of helping in the development of
technologies that would identify criminal activity.
Biometrics, the ability of computers to identify faces,
seemed to invite some considerable chances in the field
of CCTV, but the usage of the technology is not to an
extent in the town and major cities scenarios. In cases
where a still picture, captured during a criminal event,
would be compared against a database of people having
been found guilty of similar offences. The main thrust of
these developments had been in the post event detection
of crime. A live image from the street camera will
provide the comparison data. The final resolution will be
a system that works without rest or moments of
inattention and operate at the speed of human reaction to
warn operators of the presence of a known “active
criminal
entering
the
area”.
Information shared is that police previously saying that
the project is not required or not those essentials. This
http://www.ijettjournal.org
Page 955
International Journal of Engineering Trends and Technology (IJETT) - Volume4Issue4- April 2013
being the pictorial database of living criminals or all the
criminals who are active at that period of time. The
Council could display the advantages that this exchange
of information could bring without the legislation in
place there was no obligation on the police to consider
such innovative methods unless information is true.
Refinement of camera along with development of
Visions biometric software are the vital parts in the
Newham Projects . It was envisaged this totally new way
of using CCTV would take over part of the labour
intensive and stressful tasks of watching for known
individuals entering the area and comparing them against
existing knowledge of active criminals. This was not
only a innovative step for a council, but for any CCTV
operator worldwide. Thus, the recognition of criminal
faces will be guaranteed making investigation easier.
HUMAN VS. COMPUTER FACIAL SIMILARITY
RATINGS:
Criminal identification, needs applications such as it would
be very useful to have a computer system that understood
and analyze human conceptions of facial similarity; for
example, one can imagine wanting a database source
programming system that could retrieve similar faces from
a comprehensive mug shot database after a witness has
selected one face from a small initial grouping . Then
begins the interesting part where a application would be the
construction of ``electronic line-ups'': displaying a photo of
the real criminal along with a group of automaticallyselected distractor photos to test the witness's recollection.
For such an applications to be successful, the program has
to be able to encode the similarity metrics used by humans
to group faces. This shows how effectively and
successfully the current biological models are at capturing
humans' understanding of facial similarity. For testing
purpose, the computer systems were initialized with the
original neutral faces with hair. The experiments, showed
the following results.
Applications of human face scanning:






CONCLUSION:
Facial recognition is most essential method for security
purposes. The output obtained are based purely on static ,
high-quality , frontal face images. Techniques that are
used for different dimension of lighting variations ,facial
rotation, occlusion, and viewpoint changes for a purpose
of developing better facial features. The neurophysiologic
proof is sufficiently ambiguous to allow several plausible
models, with the addition of pre-processing and postprocessing steps , every model can be adjusted to fit
within the available proof .Facial recognition technology
isn't at 100% accurate as that of the other types of
biometrics because people can easily change their
appearance with beards, sunglasses, weight gain or loss
and by natural aging.
REFERENCES:
1. The results of correlation among the computer
systems and both the systems are correlated equally well
within the human ratings of similarity in the condition
with hair, whereas the graph matching system does
significantly better in the hair-free condition.
2. Both systems has the confidence to measure agreed
with human distinctiveness ratings.
3. The correlation among the human memory
performance and the PCA system is as strong in the NN
condition , which falls dramatically in the NE condition.
The graph matching system correlates about equally well
in both conditions.
ISSN: 2231-5381
Security check in airports.
Criminal identification.
Medical purposes.
Person’s identification in confidential purposes.
Capturing images.
Property privacy.
http://www.ijettjournal.org



S J Marshall, G T Reid, S J Powell, J F Towers and P
J Wells, Data capture techniques for 3-D facial
imaging, Computer Vision and Image Processing, ed.
A N Barrett, 248-275, Chapman and Hall (1991).
I D Reid, J M Brady, A McIvor, S J Marshall, I B
Knight and R C Rixon, Range vision and the Oxford
AGV, Proc. Image Processing '89 (1989), 51-72.
G T Reid, S J Marshall, R C Rixon and H Stewart, A
laser scanning camera for range data acquisition, J.
Phys. D: Appl. Phys. 21 (1988), S1-S3.
Page 956