Inter
national
J
our
nal
of
Robotics
and
A
utomation
(IJRA)
V
ol.
10,
No.
2,
June
2021,
pp.
114
122
ISSN:
2089-4856,
DOI:
10.11591/ijra.v10i2.pp114-122
r
114
Detection
of
duplicate
and
non-face
images
in
the
eRecruitment
applications
using
machine
lear
ning
techniques
Manjunath
K.
E.
1
,
Y
ogeen
S.
Honna
v
ar
2
,
Rak
esh
Pritmani
3
,
Sethuraman
K.
4
Computers
and
Information
Group,
U.
R.
Rao
Satellite
Centre
(URSC),
Indian
Space
Research
Or
g
anisation
(ISR
O),
Bang
alore,
India
Article
Inf
o
Article
history:
Recei
v
ed
Sep
26,
2020
Re
vised
Dec
1,
2020
Accepted
Feb
20,
2021
K
eyw
ords:
F
ace
detection
Haar
cascade
classifier
Histogram
Opencv
T
emplate
matching
ABSTRA
CT
The
objecti
v
e
of
this
w
ork
is
to
de
v
elop
methodologies
to
detect,
and
report
the
non-
compliant
images
with
respect
to
indian
space
research
or
g
anisation
(ISR
O)
recruit-
ment
requirements.
The
recruitment
softw
are
hosted
at
U.
R.
rao
satellite
centre
(URSC)
is
responsible
for
handling
recruitment
acti
vities
of
ISR
O.
Lar
ge
number
of
online
applications
are
recei
v
ed
for
each
post
adv
ertised.
In
man
y
cases,
it
is
observ
ed
that
the
candidates
are
uploading
either
wrong
or
non-compliant
images
of
the
required
documents.
By
non-compliant
images,
we
mean
images
which
do
not
ha
v
e
f
aces
or
there
is
not
enough
clarity
in
the
f
aces
present
in
the
images
uploaded.
In
this
w
ork,
we
attempt
to
address
tw
o
specific
problems
namely:
1)
T
o
recognise
image
uploaded
to
recruitment
portal
contains
a
human
f
ace
or
not.
This
is
addressed
using
a
f
ace
detection
algorithm.
2)
T
o
check
whether
images
uploaded
by
tw
o
or
more
applica-
tions
are
same
or
not.
This
is
achie
v
ed
by
using
machine
learning
(ML)
algorithms
to
generate
similarity
score
between
tw
o
images,
and
then
identify
the
duplicate
images.
Screening
of
v
alid
applications
becomes
v
ery
challenging
as
the
v
erification
of
such
images
using
a
manual
process
is
v
ery
time
consuming
and
requires
lar
ge
human
ef-
forts.
Hence,
we
propose
no
v
el
ML
techniques
to
determine
duplicate
and
non-f
ace
images
in
the
applications
recei
v
ed
by
the
recruitment
portal.
This
is
an
open
access
article
under
the
CC
BY
-SA
license
.
Corresponding
A
uthor:
Manjunath
K.
E.
Computers
and
Information
Group
U.
R.
Rao
Satellite
Centre
(URSC)
Indian
Space
Research
Or
g
anisation
(ISR
O)
Bang
alore,
560017-India
Email:
manjuk
e@ursc.go
v
.in
1.
INTR
ODUCTION
Computers
and
information
group
(CIG)
of
U.
R.
rao
sat
ellite
centre
(URSC)
is
in
v
olv
ed
in
de
v
elop-
ment,
customization,
and
management
of
the
softw
are
used
for
recruitment
acti
vities
of
indian
space
research
or
g
anisation
(ISR
O)
[1],
[2].
Recruitment
is
the
process
of
sourcing,
screening,
and
selecting
the
candidates
for
a
v
acanc
y
within
an
or
g
anization.
Each
year
se
v
eral
adv
ertisements
are
released,
and
fe
w
lakhs
of
applications
are
recei
v
ed
per
year
.
Scree
n
i
ng
and
processing
of
such
a
huge
v
olume
of
applications
manually
will
not
only
require
lar
ge
human
ef
forts
b
ut
also
might
lead
to
inconsistent
results.
Automation
is
the
only
solution
to
reduce
the
b
urden
from
such
repetiti
v
e
tasks.
Based
on
the
e
xpertise
g
ained
o
v
er
the
years,
certain
things
which
can
be
J
ournal
homepage:
http://ijr
a.iaescor
e
.com
Evaluation Warning : The document was created with Spire.PDF for Python.
Int
J
Rob
&
Autom
ISSN:
2089-4856
r
115
generalized
as
set
of
rules
are
already
automated.
In
addition
to
these
rule
based
automations,
in
this
w
ork,
we
w
ould
to
e
xplore
certain
image
processing
techniques
using
machine
learning
(ML)
algorithms
for
increased
automation
of
recruitment
acti
vities.
In
this
w
ork,
we
attempt
to
address
tw
o
specific
problems
namely
:
1)
T
o
recognise
image
uploaded
to
recruitment
portal
contains
a
human
f
ace
or
not.
W
e
propose
to
solv
e
this
problem
using
Haar
cascade
classifiers
based
f
ace
detection
algorithm.
2)
T
o
check
whether
images
uploaded
by
tw
o
or
more
applications
are
same
or
not.
W
e
propose
to
solv
e
this
problem
using
image
similarity
detect
ion
algorithm
based
on
certain
ML
techni
q
ue
s.
The
f
ace
detection
algorithms
w
ork
based
on
the
f
acial
features
such
as
spacing
of
the
e
yes,
bridge
of
the
nose,
the
contour
of
the
lips,
ears,
and
chin.
F
ace
detection
has
numerous
applications
in
security
(authentication
and
authorization),
defense,
mark
eting,
healthcare,
hospitality
,
f
ace
detection,
lip
reading,
and
auto-focus.
The
rest
of
the
paper
is
or
g
anized
is
being
as:
Section
2
pro
vides
brief
literature
surv
e
y
.
Section
3
describes
the
de
v
elopment
and
e
v
aluation
of
f
ace
detection
system
for
screening
of
e-recruitment
applications.
Section
4
dis
cusses
the
de
v
elopment
of
similarity
detection
system.
Section
5
summary
and
fut
ure
w
ork
change
to
conclusion.
2.
RELA
TED
W
ORK
The
research
in
f
ace
detection
and
recognition
is
v
ery
acti
v
ely
pursued
o
v
er
last
se
v
eral
dec
ades.
There
ha
v
e
been
significant
number
of
w
orks
reported
in
this
area.
Only
v
ery
fe
w
notable
w
orks
among
them
are
described
here.
Some
of
the
literature
surv
e
ys
on
the
f
ace
detection
and
recognition
is
being
as.
In
2003,
Le
wis
et
al
.
[3]
ha
v
e
presented
a
detailed
re
vie
w
on
the
psychological
e
vidence
about
the
process
of
f
ace
detection
in
brain.
It
is
sho
wn
that
with
the
use
of
f
ace
recognition
systems,
it
is
possible
to
identify
or
check
the
identity
of
indi
viduals
in
a
matter
of
fe
w
seconds.
In
2009,
Jafri
et
al
.
[4]
ha
v
e
presented
an
o
v
ervie
w
of
v
arious
f
ace
recognition
techniques.
The
benefits
and
limitations
of
dif
ferent
f
ace
recognition
algorithms
are
e
xamined.
The
applications
and
dif
ficulties
in
v
olv
ed
in
each
of
these
techniques
are
described.
In
2010,
De
gtyare
v
et
al
.
[5]
ha
v
e
proposed
set
of
parameters
for
f
ace
detection
algorithms
to
e
v
aluate
their
qualities
and
perform
objecti
v
e
comparisons,
and
to
determine
the
current
state
of
the
art
f
ace
detection
al-
gorithm.
The
y
ha
v
e
compared
se
v
en
f
ace
detection
algorithms
and
the
results
of
their
comparison
are
reported.
In
2010,
Zhang
et
al
.
[6]
ha
v
e
surv
e
yed
the
recent
adv
ances
in
f
ace
detection
for
pre
vious
decade
with
an
hope
see
better
algorithms
de
v
eloped
in
future
to
solv
e
the
problem
of
f
ace
detection.
The
y
ha
v
e
surv
e
yed
v
arious
techniques
according
to
the
w
ay
features
are
e
xtracted
and
type
of
learning
algorithms
emplo
yed.
In
2013,
Roomi
et
al
.
[7]
ha
v
e
presented
a
surv
e
y
of
v
arious
f
ace
recognition
w
orks
reported
in
the
past
decade,
mainly
focusing
on
the
ones
which
were
not
reported
in
other
similar
surv
e
ys.
Further
,
the
y
ha
v
e
cate
gorized
them
into
meaningful
approaches
such
as
appearance
based,
feature
based,
and
soft
computing
based.
A
comparati
v
e
study
of
merits
and
demerits
of
these
approaches
is
also
presented.
In
2015,
F
arf
ade
et
al
.
[8]
ha
v
e
proposed
a
deep
dense
face
detector
method
for
multi-vie
w
f
ace
detection.
The
proposed
method
does
not
require
pose/landm
ark
annotation
and
is
able
to
detect
f
aces
in
a
wide
range
of
orientations
using
a
single
model
based
on
deep
con
v
olutional
neural
netw
orks
with
minimal
comple
xity
.
In
2018,
Hua
et
al
.
[9]
ha
v
e
presented
joint
optimal
solution
for
addressing
f
ace
representation
and
matching
problems
in
f
ace
v
erification
task
using
a
unified
frame
w
ork.
A
second-order
f
ace
representa-
tion
method
for
f
ace
pair
and
a
unified
f
ace
v
erification
frame
w
ork,
in
which
the
feature
e
xtrac
tors
and
the
subsequent
binary
classification
model
design
are
made
to
select
fle
xibly
,
is
presented.
In
2020,
K
ortli
et
al
.
[10]
ha
v
e
presented
a
surv
e
y
of
some
of
the
well-kno
wn
theories
and
algorithms
used
in
f
ace
recognition.
A
detailed
comparison
in
terms
of
rob
ustness,
accurac
y
,
comple
xity
,
and
discrimi-
nation,
of
all
these
dif
ferent
techniques
is
reported.
An
o
v
ervie
w
of
the
most
comm
on
l
y
used
databases
for
both
supervised
and
unsupervised
learning
is
gi
v
en.
Frischholz
has
consolidated
all
useful
information
on
f
ace
detection
and
recognition
problems
in
[11].
It
pro
vides
appropriate
links
to
v
arious
softw
ares,
datasets,
algorithms,
selected
publications,
and
other
resources
related
to
f
ace
detection
and
recognition
problems.
There
are
fe
w
studie
s
e
xploring
the
use
of
artificial
intelligence
(AI)
techniques
for
recruitment
appli-
cations
such
as
screening
the
candidates,
establishment
of
relationships,
taking
unbiased
decisions
and
sched-
ules,
and
appli
cant’
s
social
media
communications.
Some
of
the
w
orks
e
xploring
AI
techniques
for
recruitment
acti
vities
is
being
as.
In
2018,
Upadh
yay
et
al
.
[12]
ha
v
e
re
vie
wed
the
applications
of
AI
tools
in
the
hiring
Detection
of
duplicate
and
non-face
ima
g
es
in
eRecruitment
applications
(Manjunath
K.
E.)
Evaluation Warning : The document was created with Spire.PDF for Python.
116
r
ISSN:
2089-4856
process
and
its
practical
implications.
The
y
ha
v
e
highlighted
the
strate
gic
shift
in
recruitment
industry
caused
due
to
the
adoption
of
AI
in
the
recruitment
process.
It
is
found
that
the
application
of
AI
for
managing
the
recruitment
process
is
leading
to
ef
ficienc
y
as
well
as
qualitati
v
e
g
ains
for
both
clients
and
candidates.
In
2019,
Albert
[13]
has
in
v
estig
ated
the
use
of
AI
tools
such
as
chatbots,
screening
softw
are,
and
task
automation,
in
the
recruitment
and
selection
of
candidates
by
the
companies.
On
a
similar
lines,
W
einert
et
al
.
[14]
ha
v
e
also
e
xamined
the
use
of
AI
techniques
for
selection
and
assessment
of
human
resources
by
the
companies,
and
v
arious
challenges
in
v
olv
ed
it.
In
2019,
Na
w
az
[15]
has
e
xplored
the
application
of
f
ace
detection
for
recr
uitment
process.
He
has
demonstrated
the
us
e
of
principal
component
analys
is
techniques
to
detect
duplicate
f
aces
and
thereby
enabling
the
detection
of
duplicate
applications.
In
2019,
Na
w
az
[16]
has
e
xamined
the
use
of
AI
techniques
on
the
recruitment
ef
fecti
v
eness
of
the
softw
are
companies.
The
study
uses
a
data-set
containing
a
structured
questionnaire
from
100
human
resource
professionals.
In
2019,
Esch
et
al
.
[17]
ha
v
e
w
ork
ed
on
ho
w
the
potential
candidates
re
g
ard
the
use
of
AI
in
the
recruitment
process
and
is
there
an
y
influence
on
the
lik
elihood
of
applying
for
a
job
by
potential
candidates
due
to
use
of
AI
in
recruitment.
The
y
sho
w
that
the
no
v
elty
f
actor
of
using
AI
in
the
recruitment
process,
mediates
and
further
positi
v
ely
influences
job
application
lik
elihood.
3.
F
A
CE
DETECTION
SYSTEM
FOR
SCREENING
OF
APPLICA
TIONS
Figure
1
sho
ws
the
block
diagram
of
complete
f
ace
detection
system
implemented
by
us.
A
photo
uploaded
by
an
applicant
will
be
fetched
and
fed
as
input
to
f
ace
detection
algorithm.
If
a
f
ace
is
detected
by
the
f
ace
detection
algorithm,
then
the
application
will
be
accepted.
If
a
f
ace
is
not
detected
by
the
f
ace
detection
algorithm
then
that
photo
will
be
added
to
the
list
of
ima
g
es
that
ha
v
e
to
be
manually
inspected.
The
list
of
such
images
is
made
a
v
ailabl
e
on
the
screening
portal
with
a
pro
vision
for
screening
personnel
either
to
accept
or
reject
such
applications.
The
screening
personnel
will
manually
inspect
and
accept
the
application
if
the
photo
is
proper
or
else
reject
the
application.
P
hot
o U
pl
oaded
by an
A
ppl
ic
ant
Add the
ima
ge t
o the l
ist of phot
os
t
o be Manua
lly i
ns
pe
ct
ed
Acce
pt t
he
Appli
ca
tion
Run Fac
e
D
e
tec
ti
on
Algorit
hm
Is f
a
ce
de
t
e
ct
e
d?
Y
e
s
No
I
s face
de
t
e
ct
e
d?
Inspec
t Manua
lly
Y
es
R
e
je
ct
the
Appli
cat
ion
Figure
1.
Block
diagram
of
f
ace
detection
system
3.1.
F
ace
detection
algorithm
F
ace
detection
is
an
image
processing
technique
for
identifying
human
f
aces
in
images
and
videos.
It
is
the
psychological
process
with
which
humans
locate
and
attend
to
f
aces
in
a
visual
scene
[3].
F
ace
detection
is
a
specific
case
of
object
detection,
where
f
ace
becomes
the
object
to
be
detected.
The
task
of
object
detection
is
to
find
the
locations
and
sizes
of
all
objects
in
an
im
age
that
belong
to
a
gi
v
en
class.
In
this
w
ork,
we
ha
v
e
w
ork
ed
on
f
ace
detection
using
a
haar
cas
cade
classifiers.
The
f
ace
detection
using
Haar
feature-based
cascade
classifiers
is
a
machine
learning
based
approach
where
a
cascade
function
is
trained
using
lar
ge
number
of
positi
v
e
and
ne
g
ati
v
e
images
[18].
The
trained
cascade
functi
on
is
used
to
detect
similar
objects
in
other
images.
Haar
features
are
lik
e
con
v
oluctional
k
ernel,
where
each
feature
is
a
single
v
alue
obtained
by
subtracting
sum
of
pix
els
under
white
rectangle
from
sum
of
pix
els
under
black
rectangle
[19].
The
haar
features
are
computed
Int
J
Rob
&
Autom,
V
ol.
10,
No.
2,
June
2021
:
114
–
122
Evaluation Warning : The document was created with Spire.PDF for Python.
Int
J
Rob
&
Autom
ISSN:
2089-4856
r
117
by
finding
the
sum
of
pix
els
under
white
and
bla
ck
rectangles.
The
calculation
of
sum
of
pix
els
is
simplified
using
inte
gral
images.
Lar
ge
number
of
features
are
computed
using
all
po
s
sibles
sizes
and
locations
of
each
k
ernel.
The
four
Haar
features
namely
are:
a)
Edge
feature,
b)
Line
feature,
c)
F
our
-rectangle
feature.
F
igure
2
sho
ws
the
v
arious
types
of
Haar
features
for
f
ace.
The
edge
features
seems
to
focus
on
the
property
that
the
re
gion
of
the
e
yes
is
often
dark
er
than
the
re
gion
of
the
nose
and
cheeks,
while
the
line
features
focus
on
the
property
that
the
e
yes
are
dark
er
than
the
bridge
of
the
nose.
These
features
are
detected
only
when
the
windo
w
is
applied
on
the
f
ace
re
gion,
and
the
windo
ws
applyi
ng
on
cheeks
or
an
y
other
part
of
the
image
become
irrele
v
ant.
Each
and
e
v
ery
feature
is
applied
on
all
the
training
images.
F
or
each
feature,
it
finds
the
best
threshold
which
will
classify
the
f
aces
into
positi
v
e
and
ne
g
ati
v
e
classes.
The
features
with
minimum
error
rate
are
selected.
These
features
indicate
that
the
y
are
the
features
that
best
classifies
the
f
ace
and
non-f
ace
images.
W
e
ha
v
e
used
the
pre-trained
Haar
cascade
classifier
model
pro
vided
by
Opencv
[19]
library
.
Figure
2.
Illustration
of
v
arious
types
of
Haar
features
for
f
ace
detection
(Courtesy:
Figure
tak
en
from
[20])
3.2.
Ev
aluation
of
face
detection
algorithm
W
e
ha
v
e
ran
the
f
ace
detection
algorithm
for
some
of
our
selected
recruitment
adv
ertisements.
T
able
1
sho
ws
the
e
v
aluation
statistics
of
f
ace
detection
system.
From
the
column
4
,
it
can
be
seen
that
some
of
the
v
alid
photos
are
also
detected
as
in
v
alid
photos.
Hence,
we
can
not
blindly
use
the
output
of
f
ace
detection
algorithm
as
it
is.
The
list
of
suspected
in
v
alid
photos
ha
v
e
to
be
inspected
manually
and
actual
in
v
alid
photos
ha
v
e
to
be
determined.
This
mak
es
the
f
ac
e
detection
system
semi-automatic.
Although
this
system
can
not
replace
the
human
interv
ention
completely
,
b
ut
it
drastically
reduces
the
human
ef
fort
in
v
olv
ed
in
screening
of
recruitment
applications.
S
ixth
column
in
T
able
1
sho
ws
the
%
r
eduction
in
the
manual
ef
fort
for
scr
eening
applications.
The
a
v
erage
reduction
is
98.55%,
which
indicates
only
1.45%
of
the
manual
ef
fort
required
for
performing
the
screening
using
f
ace
detection
system.
This
is
a
v
ery
drastic
reduction
in
the
manual
ef
fort.
F
or
e
xample,
in
case
of
serial
no.
1
(second
ro
w),
the
use
of
f
ace
detection
system
has
reduced
the
number
of
applications
to
be
screened
from
4145
to
79
.
Lik
e
wise
the
reduction
is
from
30008
to
304
for
serial
no.
6
(se
v
enth
ro
w).
The
last
column
pro
vides
the
f
ace
detection
accurac
y
.
The
a
v
erage
f
ace
detection
accurac
y
is
found
to
be
76.41%,
which
is
reasonably
a
good
v
alue.
This
approach
w
ould
not
only
reduce
the
costs
in
v
olv
ed
in
recruitment
acti
vities
b
ut
also
promises
more
consistent
results,
and
requires
v
ery
less
time
compared
to
humans.
This
approach
will
not
gi
v
e
an
y
chance
to
miss
out
an
y
of
the
applications
with
v
alid
photos
as
an
y
rejection
will
al
w
ays
ha
v
e
to
be
done
by
humans.
Figure
3
sho
ws
fe
w
suspected
in
v
alid
photos
detected
by
f
ace
detection
a
lgorithm.
It
is
v
ery
s
u
r
prising
to
see
v
arious
dif
ferent
kinds
of
photos
uploaded
by
the
candidates
along-with
their
applications.
In
v
alid
photos
v
ary
from
animations,
signatures,
marks
cards,
snapshot
of
mobiles,
whatsapp
images,
some
random
image
tak
en
from
internet,
and
some
random
photo
click
ed
using
mobiles.
Due
to
data
confidentiality
issues,
we
ha
v
e
sho
wn
only
the
generic
images
in
Figure
3.
Ho
we
v
er
,
there
are
se
v
eral
v
ariety
of
images
such
as
certificates,
grade
cards,
photo
images,
(which
are
of
restricted
nature
and
can
not
be
published)
that
were
also
classified
as
in
v
alid
images
by
the
algorithm.
Fe
w
such
e
xamples
include
1)
f
aces
in
the
image
are
completel
y
co
v
ered
by
hairs
such
that
only
one
side
of
the
f
ace
is
visible,
2)
photos
that
are
captured
using
the
head
co
v
ered
with
a
cap
or
a
turban
such
that
part
of
the
forehead
is
not
visible,
3)
photos
are
tak
en
such
that
part
of
the
forehead,
cheeks
and
chin
are
not
visible,
and
4)
photos
with
goggles
Detection
of
duplicate
and
non-face
ima
g
es
in
eRecruitment
applications
(Manjunath
K.
E.)
Evaluation Warning : The document was created with Spire.PDF for Python.
118
r
ISSN:
2089-4856
co
v
ering
their
e
yes.
Hence,
in
some
of
the
cases
the
f
ace
detection
algorithm
has
f
ailed
to
detect
a
human
f
ace
due
to
follo
wing
reasons.
1)
If
the
photo
is
tak
en
by
wearing
a
spectacle.
I
n
this
case,
the
algorithm
f
ails
to
detect
t
h
e
f
acial
feat
ures
such
as
spacing
of
the
e
yes,
and
the
contrasting
line
features
present
at
the
e
yebro
ws
and
e
yeball
co
v
ers
are
lost,
2)
If
an
head
cap
or
turban
is
used
such
that
certain
part
of
forehead
and
e
yebro
ws
are
co
v
ered,
and
complete
f
ace
is
not
visible.
In
this
case
also
algorithm
f
ails
to
e
xtract
all
the
f
acial
features,
3)
If
the
f
ace
is
rotated
such
that
only
one
side
of
the
f
ace
is
visible,
and
other
side
of
the
f
ace
is
either
partially
or
completely
in
visible,
then
algorithm
will
not
able
capture
all
the
required
features,
4)
If
the
resolution
of
the
image
is
too
lo
w
,
so
that
considered
windo
w
size
e
xceeds
the
photo
size.
T
able
1.
Ev
aluation
statistics
of
f
ace
detection
system
Sl
No.
T
otal
No.
of
Applications
Screened
Suspected
In-
v
alid
Photos
Count
No.
of
Correct
Photos
Detected
as
In
v
alid
Photos
No.
of
Incorrect
Photos
Detected
as
In
v
alid
Photos
%
Reduction
in
the
Manual
ef-
fort
F
ace
Detec-
tion
Accurac
y
(%)
1
4145
79
18
61
98.09
77.21
2
3706
62
12
50
98.32
80.64
3
45059
414
156
258
99.08
62.31
4
25700
237
62
175
99.07
73.83
5
10280
106
19
87
98.96
82.07
6
30008
304
51
253
98.98
83.22
7
8787
144
33
111
98.36
77.08
8
832
15
3
12
98.19
80
9
1727
27
3
24
98.43
88.88
10
869
17
7
10
98.04
58.82
A
v
erage
%
-
-
-
-
98.55
76.41
Figure
3.
Sample
in
v
alid
photos
detected
by
f
ace
detection
system
Int
J
Rob
&
Autom,
V
ol.
10,
No.
2,
June
2021
:
114
–
122
Evaluation Warning : The document was created with Spire.PDF for Python.
Int
J
Rob
&
Autom
ISSN:
2089-4856
r
119
4.
SIMILARITY
DETECTION
SYSTEM
FOR
PHO
T
OS
T
w
o
importa
n
t
techniques
for
comparison
of
images
are
1)
Comparison
of
histograms
and
2)
T
em
p
l
ate
matching.
An
histogram
is
a
graphical
representation
of
the
v
alue
distrib
ution
of
a
digital
image.
The
histogram
intersection
algorithm
w
as
proposed
by
Sw
ain
and
Ballard
in
[21].
The
histogram
intersection
does
not
require
the
accurate
separation
of
the
object
from
its
background
and
it
is
rob
ust
to
occluding
objects
in
the
fore
ground.
Histograms
are
translation
in
v
ariant,
b
ut
the
y
change
slo
wly
under
dif
ferent
vie
w
angles,
scales
and
in
presence
of
occlusions
[22].
Histogram
comparison
is
one
of
the
simplest,
f
astest
method
to
find
the
s
imilarities
in
the
images.
Here
the
assumption
is
that
a
parti
cular
type
of
picture
will
ha
v
e
a
particular
color
in
ab
undance.
F
or
e
xample,
a
picture
of
a
forest
will
ha
v
e
a
lot
of
green
color
,
a
picture
of
a
banana
will
ha
v
e
lot
of
yello
w
color
.
So,
if
tw
o
pictures
with
forests
are
being
compared
then
we
will
get
some
similarity
between
the
tw
o
histograms,
as
both
of
them
ha
v
e
lot
of
green
color
.
Further
details
on
comparison
of
histograms
can
be
found
in
[21],
[22].
T
emplate
matching
is
a
technique
in
digital
image
processing
for
finding
small
parts
of
an
image
which
match
a
template
image.
A
basic
method
of
template
matching
uses
an
image
template,
tailored
to
a
specific
feature
of
the
search
image
which
we
w
ant
to
detect.
The
cross
correlation
output
will
be
highest
at
places
where
the
image
structure
matches
the
mask
structure,
where
lar
ge
image
v
alues
get
multiplied
by
lar
ge
mask
v
alues.
As
all
possible
posi
tions
of
the
template
with
respect
to
the
search
image
are
considered,
the
position
with
the
highest
score
is
the
best
position
[23],
[24].
It
is
kno
wn
w
ork
well
with
identical
images
with
same
size
and
orientation,
to
which
our
case
mostl
y
fits
in.
Further
details
on
template
matching
can
be
found
in
[23],
[24].
In
this
study
,
we
ha
v
e
computed
the
similarity
score
using
the
combination
of
both
the
approaches
-
comparison
of
histograms
and
template
matching.
Python’
s
OpenCV
library
is
used
for
implementation.
Since,
both
of
these
methods
alone
did
not
produce
better
results,
we
ha
v
e
combined
them
using
a
weighted
combination
method.
W
e
ha
v
e
assigned
a
lo
wer
weightage
of
0.1
to
histo
gr
am
comparison
method
as
i
t
w
as
found
to
be
less
accurate
than
template
matc
hing
method.
And,
template
matc
hing
method
w
as
assigned
a
higher
weightage
of
0.9.
T
w
o
images
are
compared
and
a
similarity
score
is
returned
based
on
the
comparison.
The
similarity
score
indicates
”ho
w
similar
the
tw
o
images
being
compared
are”.
F
or
e
xample,
a
similarity
score
of
100%
w
ould
indicate
that
the
same
image
is
being
compared,
and
a
similarity
score
of
0%
w
ould
indicate
that
tw
o
images
are
totally
dif
ferent.
Each
image
in
an
adv
ertisement
will
be
compared
with
all
other
images.
This
w
ould
result
in
a
ti
me
comple
xity
of
O(
n
2
).
After
comparison
of
images,
the
algorithm
w
ould
return
a
similarity
score
ranging
from
0%
to
100%.
In
this
study
,
we
ha
v
e
considered
only
the
cases
with
similarity
score
of
100%
.
The
comparison
of
images
that
ha
v
e
returned
a
similarity
score
of
100%
w
ould
be
treated
as
similar
images
.
This
algorithm
is
computationally
v
ery
intensi
v
e
and
requires
huge
computing
resources.
F
or
one
instance
of
comparison
of
pair
of
images
on
a
Desktop
PC
(8
GB
RAM,
Intel
i7-6700
CPU
@
3.40GHz
with
8
cores,
No
Graphics
card)
took
around
one
minute.
Although
t
he
proposed
technique
is
w
orking
reasonably
well
and
has
produced
some
of
the
promising
results,
due
to
data
confidenti
ality
issues,
we
are
restricted
to
not
to
publish
an
y
of
the
images
that
are
detected
by
the
similarity
detection
system.
W
e
ha
v
e
found
that,
there
are
number
of
instances
where
the
same
candidate
has
applied
multiple
times
to
the
same
post
adv
ertised
using
the
same
photo.
In
one
such
case,
we
found
that
a
candidate
has
applied
5
times
to
the
same
post
using
the
same
photo.
5.
SUMMAR
Y
AND
FUTURE
W
ORK
In
this
w
ork,
we
ha
v
e
e
xplored
tw
o
ML
techniques-f
ace
detection
and
similarity
detection-for
aut
omat-
ing
the
screening
of
recruitment
applications.
It
is
found
that
the
use
of
f
ace
detection
system
has
drastically
reduced
(by
98.5%)
the
manual
ef
fort
required
for
screening
the
recruitment
applications.
The
detailed
analy-
sis
on
when
and
wh
y
the
f
ace
detect
ion
f
ails
is
carried
out.
The
similarity
detection
system
w
as
de
v
eloped
to
compare
tw
o
images
and
det
ermine
their
similarity
score.
Although,
the
similarity
detection
system
is
w
orking
reasonably
well
b
ut
it
is
v
ery
resource
hungry
and
requires
lar
ge
computing
infrastructure.
In
future,
v
arious
state-of-the-art
deep
le
arning
algorithms
such
as
con
v
olutional
neural
netw
orks
(CNN)
for
f
ace
detection
[25],
[26]
can
be
e
xplored
to
detect
and
eliminate
non-f
ace
images.
Instead
of
using
the
libraries
pro
vided
by
OpenCV
,
the
f
ace
detection
models
can
be
trained
using
custom
datasets
of
f
ace
and
non-f
ace
images,
and
then
these
models
can
be
used
for
performing
f
ace
detection.
One
can
also
e
xplore
the
Detection
of
duplicate
and
non-face
ima
g
es
in
eRecruitment
applications
(Manjunath
K.
E.)
Evaluation Warning : The document was created with Spire.PDF for Python.
120
r
ISSN:
2089-4856
possiblity
of
de
v
elopment
of
h
ybrid
techniques
(which
combine
outputs
of
multiple
f
ace
detection
algorithms)
for
f
ace
detection.
The
feature
mapping
techniques
can
be
e
xplored
for
b
uilding
similarity
detection
systems
for
similarity
detection
of
f
ace
images.
Sparse
coding
based
image
similarity
detection
[27]
techniques
can
be
e
xplored
for
b
uilding
similarity
detection
systems.
REFERENCES
[1]
K.
M.
Pratap,
S.
Y
.
S.
Honna
v
ar
,
R.
K
umar
,
S.
K
umar
D,
and
D.
Gurumoorth
y,
“e-Recruitment
in
ISR
O:
Adv
antages
and
Challenges,
”
in
W
orkshop
on
Computer
and
Information
T
ec
hnolo
gy
(WCIT)
,
2008.
[2]
Y
.
Honna
v
ar
,
K.
M.
Pratap,
R.
K
umar
,
S.
Ramanathan,
and
G.
N.
V
.
Prasad,
“Enabling
Digital
P
ayments
-
a
case
study,
”
in
ISR
O
Symposium
on
Computer
s
and
Information
T
ec
hnolo
gy
(ISCIT)
,
2018.
[3]
Le
wis,
M.
Be
v
an,
and
H.
D.
Ellis,
“Ho
w
we
detect
a
f
ace:
A
surv
e
y
of
psychological
e
vidence,
”
In-
ternational
J
ournal
of
Ima
ging
Systems
and
T
ec
hnolo
gy
,
v
ol.
13,
no.
1,
pp.
3-7,
September
2003,
doi:
10.1002/ima.10040.
[4]
R.
Jafri
and
H.
Arabnia,
“A
Surv
e
y
of
F
ace
Recognition
T
echniques,
”
J
ournal
of
Information
Pr
ocessing
Systems
,
v
ol.
5,
no.
2,
pp.
41-68,
2009.
doi:
10.3745/JIPS.2009.5.2.041.
[5]
N.
De
gtyare
v
and
O.
Seredin,
“Comparati
v
e
T
esting
of
F
ace
Detection
Algorithms,
”
Ima
g
e
and
Signal
Pr
ocessing
,
(Spring
er)
,
v
ol.
6134,
pp.
200-209,
2010.
doi:
10.1007/978-3-642-13681-8
24.
[6]
C.
Zhang
and
Z.
Zhang,
“A
Surv
e
y
of
Recent
Adv
ances
in
F
ace
Detection,
”
T
ec
hnical
Report
-
Micr
osoft
Resear
c
h
Publication
,
MSR-TR-2010-66,
2010.
[7]
M.
Roomi
and
M.
P
.
Beham,
“A
Re
vie
w
of
F
ace
Recognition
Methods,
”
Cir
cuits,
Systems,
and
Signal
Pr
ocessing
,
(Spring
er)
,
v
ol.
27,
no.
4,
pp
1-35,
2013.
doi:
10.1142/S0218001413560053.
[8]
S.
S.
F
arf
ade,
M.
Saberian,
and
L.
Li,
“Multi-vie
w
F
ace
Detection
Using
Deep
Con
v
olutional
Neural
Netw
orks,
”
Pr
oceedings
of
the
5th
A
CM
on
International
Confer
ence
on
Multimedia
Retrie
val
(ICMR)
,
2015,
pp.
643-650,
doi:
10.1145/2671188.2749408.
[9]
Q.
Hua,
C.
Dong,
and
F
.
Zhang,
“A
No
v
el
Approach
to
F
ace
V
erification
Based
on
Second-Order
F
ace-
P
air
Representation,
”
Hindawi
Comple
xity
(W
ile
y)
,
pp.
1-10,
2018.
doi:
10.1155/2018/2861695.
[10]
Y
.
K
ortli,
M.
Jridi,
A.
A.
F
alou,
and
M.
Atri,
“F
ace
Recognition
Systems:
A
Surv
e
y,
”
Sensor
s,
MDPI
,
v
ol.
20,
no.
2,
pp.
3-34,
2020.
doi:
10.3390/s20020342.
[11]
R.
Frischholz,
“F
ace
Detection
&
Recognition
Homepage,“
[Online].
A
v
ailable:
https://f
acedetection.com
[Accessed
Oct.
07,
2020].
[12]
A.
K.
Upadh
yay
and
K.
Khandel
w
al,
“Appl
ying
artificial
intelligence:
i
mplications
for
recruitment,
”
Str
ate
gic
HR
Re
vie
w
,
2018.
doi:
10.1108/SHR-07-2018-0051.
[13]
E.
T
.
Albert,
“AI
in
talent
acquisition:
A
re
vie
w
of
AI-applications
used
in
recruitment
and
selection,
”
Str
ate
gic
HR
Re
vie
w
,
v
ol.
18,
no.
5,
pp.
215-221,
2019.
doi:
10.1108/SHR-04-2019-0024.
[14]
S.
W
einert,
E.
G
¨
unther
,
E.
Rue
ger
-Muck,
and
G.
Raab,
“Artificial
intelligence
in
personnel
selection
and
its
influence
on
emplo
yer
attracti
v
eness,
”
Cr
oss-Cultur
al
Business
Confer
ence
,
v
ol.
15,
no.
3,
pp.
22-35,
2020.
doi:
10.1007/s00034-017-0568-8.
[15]
N.
Na
w
az,
“
Artificial
Intelligence
F
ace
Recognition
for
applicant
tracking
system,
”
International
J
ournal
of
Emer
ging
T
r
ends
in
Engineering
Resear
c
h
,
v
ol.
7,
no.
12,
pp.
895-901,
2019.
doi:
10.30534/ijeter/2019/277122019.
[16]
N.
Na
w
az,
“
Artificial
Intelligence
Is
T
ransforming
Recruitment
Ef
fecti
v
eness
in
CMMI
Le
v
el
Compa-
nies,
”
International
J
ournal
of
Advanced
T
r
ends
in
Computer
Science
and
Engineering
,
v
ol.
8,
no.
6,
pp.
3017-3021,
2019,
doi:
10.30534/ijatcse/2019/56862019.
[17]
P
.
v
.
Esch,
J.
S.
Black,
and
J.
Ferolie
,
“Mark
eting
AI
recruitment:
The
ne
xt
phase
in
job
appli
cation
and
selection,
”
Computer
s
in
Human
Behavior
,
v
ol.
90,
pp.
215-222,
2019.
doi:
10.1016/j.chb
.2018.09.009.
[18]
P
.
V
iola
and
M.
Jones,
“Rapid
Object
Detection
using
a
Boosted
Cascade
of
Simple
Features,
”
in
IEEE
Computer
Society
Confer
ence
on
Computer
V
ision
and
P
attern
Reco
gnition.
CVPR
2001
,Kauai,
HI,
USA,
v
ol.
1,
2001,
pp.
115-118,
doi:
10.1109/CVPR.2001.990517.
[19]
Opencv
Python
T
utorials.
F
ace
Detection
using
Haar
Cascades,
[Online].
A
v
ail-
able:
https://opencv-p
ython-tutroals.readthedocs.io/en/latest/p
y
tutorials/p
y
objdetect
/p
y
f
ace
detection/p
y
f
ace
detection.html.
[20]
E.
T
yanto
v
,
“F
ace
Recognition:
From
Scratch
T
o
Hatch,
[Online].
A
v
ailable:
https://en.ppt-
online.or
g/354650.
Int
J
Rob
&
Autom,
V
ol.
10,
No.
2,
June
2021
:
114
–
122
Evaluation Warning : The document was created with Spire.PDF for Python.
Int
J
Rob
&
Autom
ISSN:
2089-4856
r
121
[21]
M.
J.
Sw
ain
and
D.
H.
Ballard,
“Color
inde
xing,
”
International
J
ournal
of
Computer
V
ision
,
v
ol.
7,
no.
1,
pp.
11-32,
1991,
doi:
10.1007/BF00130487.
[22]
M.
P
atacchiola,
“The
Simplest
Classifier:
Histogram
Comparison,
”
[Online].
A
v
ailable:
https://mpatacchiola.github
.io/blog/2016/11/12/the-simplest-classifier
-histogram-intersection.html.
[23]
R.
Brunelli,
“
T
emplate
Matc
hing
T
ec
hniques
in
Computer
V
ision:
Theory
and
Pr
actice
,
”
John
W
ile
y
Sons,
2009.
[24]
C.
T
.
Y
uen,
M.
Rizon,
W
.
S.
San,
and
T
.
C.
Seong,
“F
acial
Features
for
T
emplate
Matching
Based
F
ace
Recognition,
”
American
J
ournal
of
Engineering
and
Applied
Sciences
,
v
ol.
3,
no.
1,
pp.
899-903,
2010.
[25]
M.
Cos
¸
kun,
A.
Uc
¸
ar
,
¨
O.
Y
ildirim,
and
Y
.
Demir,
“F
ace
recognition
based
on
con
v
olutional
neural
net-
w
ork,
”
2017
International
Confer
ence
on
Modern
Electrical
and
Ener
gy
Systems
(MEES)
,
Kremenchuk,
Ukraine,
2017,
pp.
376-379,
doi:
10.1109/MEES.2017.8248937.
[26]
Kalino
vskii,
Ilya,
and
V
.
Spitsyn,
“Compact
con
v
olutional
ne
u
r
al
netw
ork
cascade
for
f
ace
detection,
”
arXiv
pr
eprint
arXiv:1508.01292
,
2015.
[27]
L.
Xi-dao,
X.
Y
u-xiang,
Z.
Li-li,
Z.
Xin,
L.
Chen,
and
H.
Jingmeng,
“
An
Image
Similarity
Acceleration
Detection
Algorithm
Based
on
Sparse
Coding,
”
Mathematical
Pr
oblems
in
Engineering
,
2018,
pp.
1-9,
doi:10.1155/2018/1917421.
BIOGRAPHIES
OF
A
UTHORS
Dr
.
Manjunath
K.
E.
recei
v
ed
his
BE
in
Computer
Science
from
SIT
T
umakuru,
MS
from
IIT
Kharagpur
,
and
Ph.D
from
IIIT
Bang
alore
in
the
area
of
speech
processing.
He
is
currently
w
orking
as
scientist
at
U.
R.
Rao
Satellite
C
entre,
Bang
alore.
He
has
w
ork
ed
as
softw
are
engineer
in
reputed
softw
are
firms
for
about
2.5
years
before
joining
ISR
O
in
2016.
His
research
interest
s
include
speech
recognition,
machine
learning,
and
automation
of
data
center
operations.
He
has
published
o
v
er
20
research
articles
in
v
arious
conferences
and
journals.
He
has
also
authored
te
xt
book
on
speech
recognition
using
machine
learning
algorithms.
Y
ogeen
S.
Honna
v
ar
recei
v
ed
his
BE
in
Computer
Science
from
Karnatak
Uni
v
ersity
in
2000,
M.T
ech
in
Information
T
echnology
from
Karnataka
Uni
v
ersity
in
2012.
He
joined
U.
R.
Rao
Satellite
Centre
in
March
2004
and
currently
he
is
Head,
Information
Security
Section,
Computers
and
Infor
-
mation
Group.
He
has
published
research
articles
in
national
conferences
and
seminars.
His
areas
of
w
ork
include
e-Recruitment
for
ISR
O,
Cybersecurity
.
Rak
esh
Pritmani
recei
v
ed
his
M.T
ech
de
gree
in
Computer
Science
in
1998
and
M.Sc.
Electron-
ics
de
gree
in
1992
from
De
vi
Ahil
ya
Uni
v
ersity
,
Indore.
He
joined
U.
R.
Rao
Satellite
Centre
in
December
2000
and
currently
he
is
Head,
Central
C
omputer
Systems
Di
vision,
Computers
and
In-
formation
Group.
Before
joining
URSC
he
has
w
ork
ed
as
Lecturer
in
School
of
Electronics,
De
vi
Ahilya
Uni
v
ersity
for
a
period
of
7
years.
He
has
played
k
e
y
role
in
implementation
of
T
eamcenter
PLM
solution
at
U.
R.
Rao
Satellite
Centre.
He
has
published
research
articles
in
national
confer
-
ences
and
semina
rs.
His
areas
of
w
ork
include
High
Performance
Computing
Systems,
Product
Life
Cycle
Management
systems.
Rak
esh
K
umar
recei
v
ed
his
B.Sc
(Engg)
de
gree
in
Computer
Science
and
Engg
in
1993
from
V
inoba
Bha
v
e
Uni
v
ersity
,
Hazaribag
and
M.
T
ech
de
gree
in
Computer
Science
and
Engg
from
VTU,
Belg
aum
in
2009.
He
joined
U.
R.
Rao
Satellite
Centre,
Beng
aluru
in
1996.
He
is
heading
the
Computers
and
Information
Group
at
URSC.
He
is
responsible
for
managing
secured
interne
t
services
at
URSC
using
layered
c
yber
security
implementation
and
for
managing
v
arious
application
softw
are/f
acilities
used
at
URSC
f
acilitated
by
CIG
such
as
e-Recruitment
for
ISR
O,
web
based
w
orkflo
w
softw
are,
Messaging
System
of
URSC.
He
has
published
about
10
technical
papers
in
v
arious
National
and
International
conferences/journals.
Detection
of
duplicate
and
non-face
ima
g
es
in
eRecruitment
applications
(Manjunath
K.
E.)
Evaluation Warning : The document was created with Spire.PDF for Python.
122
r
ISSN:
2089-4856
Sethuraman
K.
is
currently
Deputy
Director
,
Management
and
Information
Systems
Area,
U.
R.
Rao
Satellite
Centre.
Prior
to
this,
he
w
as
functioning
as
Director
,
Satellite
Communication
and
Na
vig
a-
tion
Programme
Of
fice
in
ISR
O
Headquarters,
Bang
al
ore.
Mr
.
Sethuraman
w
as
serving
in
the
satel-
lite
communication
programme
of
fice,
since
2000,
in
v
arious
capacities.
His
contrib
utions
included
Frequenc
y
management
functions,
INSA
T/GSA
T
satelite
transponder
management,
SA
TCOM
pol-
ic
y
implementation
and
implement
ing
societal
applications
lik
e
tele-education,
telemedicine,
V
illage
Resource
Centre
etc.
Prior
to
joini
ng
Satellite
Communication
Programme
Of
fice,
he
w
as
w
orking
in
the
areas
of
satellite
mission
operations
with
special
emphasis
on
real-time
softw
are
management
functions
at
INSA
T
-Master
Control
F
acility
at
Hassan,
for
o
v
er
15
years.
Int
J
Rob
&
Autom,
V
ol.
10,
No.
2,
June
2021
:
114
–
122
Evaluation Warning : The document was created with Spire.PDF for Python.