ATTENDANCE MANAGEMENT SYSTEM

CS4089 Project

Midterm Report

Krishna Sai Gullapally, Prashanth N, MS Bhanu Prakash, Sai Bhava Teja C

Guided By: Dr. Lijiya A.

October 4, 2017

1 Introduction ods like Histogram Equalization, Con-

trast Stretching and Binarization.
Maintaining and recording attendance of Pattern recognition is the process of ex-
daily wage workers at construction sites tracting a pattern from a given image us-
has always been a difficult task for the ing methods like Segmentation, Normal-
organizations. Over the years manual ization and Gabor filters.
attendance system has been carried out Pattern Matching is finding the highest
across most of the organizations. To match of the pattern in the database.
overcome the inefficiencies in the man- There are two matching techniques
ual recording of attendance, organiza- which are Identification and Verification.
tions have been testing automated so- Identification is one to many matching
lutions using biometric techniques like where it compares the extracted pattern
fingerprint, face recognition, voice recog- with all the entries in the database. Veri-
nition etc., However these biometric so- fication is one to one which is comparing
lutions tends to give inaccurate results the extracted pattern with a particular
in real life deployment. Since unimodal entry in the database.
biometric techniques have to contend
with a variety of problems such as noisy
data, spoof attacks and unacceptable er-
2 Problem Statement
ror rates. Some of these limitations can
To design and build an efficient auto-
be solved by using multi-level biometric
matic attendance management system
systems by reducing the threshold of in-
based on biometrics for daily wage work-
dividual biometric techniques.
ers in the construction sites.
The topic of our project is At-
tendance Management System for
Daily Wage Workers. The solution we 3 Literature Survey
have opted for is based on Bio-metrics.
This project involves knowledge in Image A number of related works exist on the
Processing, Database Management Sys- application of different methods and
tems and Software Design. principles to effectively monitor the
Image Processing involves techniques attendance . Solutions can either be bio-
like Image Enhancement, Pattern Recog- metric or non-biometric. The solutions
nition and Pattern Matching. based on non biometric which include
Image enhancement is the process of ad- RFID-based and GSM-GPRS based
justing digital images so that the re- attendance management systems suffer
sults are more suitable for further image a lot of problems and are impractical to
analysis and it includes different meth- implement in real world scenarios. So,

1
the study was constrained to biometrics. to one of the three families: Optical
Detailed study on biometric methods sensors,Solid-state sensors,Ultrasound
such as fingerprint, facial and iris recog- sensors.
nition is done.From the study, a general New sensing techniques such as mul-
block diagram of the various biometric tispectral imaging and 3D touchless
systems was conceived in the following acquisition are being developed to
diagram. overcome some of the drawbacks of the
current fingerprint scanners.
Image Enhancement[2]
Enhancement algorithm improves the
clarity of ridge and valley structures
in the fingerprint images. Implications
of poor quality image: A significant
number of spurious minutiae may be
created, A large percent of genuine
minutiae may be ignored, and Large
errors in their localization (position and
orientation) may be introduced. Follow-
ing steps constitute the enhancement
algorithm: 1. Normalization: An input
image is normalized, so that the ridges
and valleys are easily distinguished.
2. Local Orientation Estimation: From
the normalized fingerprint image orien-
tation of ridges are calculated in each
block of desired size.
FINGERPRINT RECOGNI-
3. Frequency Estimation: Using the
TION SYSTEM
normalized image and the Orientation
Fingerprints are considered to be the
image the frequency in each of the
best and fastest method for biometric
blocks of the image is calculated.
identification. They are secure to
4. Filtering: A bank of Gabor filter
use, unique for every person and does
tuned to local ridge orientation and
not change in ones lifetime. Besides
frequency is used to filter the image to
these, implementation of fingerprint
clearly separate out ridges and valleys
recognition system is cheap,easy and
and thus reduce the probability of
accurate up to satisfiability.Fingerprint
spurious minutiae.
recognition[1] has been widely used in
both forensic and civilian applications.
Compared with other biometrics fea-
tures , fingerprint-based biometrics is
the most proven technique and has the
largest market shares.Not only it is
faster than other techniques but also the
energy consumption by such systems is
too less.
Hardware
The quality of a fingerprint scanner, the
size of its sensing area and the resolution
can heavily influence the performance
of a fingerprint recognition algorithm. Fingerprint Segmentation
Almost all the existing sensors belong The most important step in minutiae-

2
based matching is to extract reliablyFingerprint matching techniques
the minutiae points from the input The large number of approaches to
image. This extraction depends on thefingerprint matching can be coarsely
efficiency of the detection of ridges and
classified into three families.
valleys in the fingerprint segmentation
Correlation-based matching: Two fin-
step. Therefore an accurate segmenta-gerprint images are superimposed and
tion of fingerprint images produces an
the correlation between corresponding
effective result in automatic fingerprint
pixels is computed for different align-
recognition system. ments (e.g. various displacements and
Fingerprint image consist of two parts,
rotations).
foreground part that is the essentialMinutiae-based matching: This is the
area for extracting minutiae points, most popular and widely used technique,
and background part that is the noisybeing the basis of the fingerprint com-
area. Segmentation of the fingerprintparison made by fingerprint examiners.
image is to decide which part from the
Minutiae are extracted from the two
image is associated with the foreground
fingerprints and stored as sets of points
and which part is associated with thein the two- dimensional plane. Minutiae-
background. Due to the nature of
based matching essentially consists of
fingerprint image and the presence offinding the alignment between the
noise, the decision for separation these
template and the input minutiae sets
two regions is critical. that results in the maximum number of
Feature extraction minutiae pairings.
When analyzed at the global level, Pattern-based matching: Pattern based
the fingerprint pattern exhibits one algorithms compare the basic fingerprint
or more regions where the ridge linespatterns (arch, whorl, and loop) between
assume distinctive shapes. These re- a previously stored template and a can-
gions (called singularities or singular
didate fingerprint. This requires that
regions) may be classified into threethe images be aligned in the same orien-
typologies: loop, delta, and whorl. At
tation. To do this, the algorithm finds
the local level, other important features,
a central point in the fingerprint image
called minutiae can be found in the and centers on that. In a pattern-based
fingerprint patterns. Minutia refers to
algorithm, the template contains the
the various ways in which the ridges type, size, and orientation of patterns
can be discontinuous. For example,
within the aligned fingerprint image.
a ridge can abruptly come to an end The candidate fingerprint image is
(termination), or can divide into twographically compared with the template
ridges (bifurcation). Minutia extraction
to determine the degree to which they
Algorithms are then employed to extract
Match.
the minutiae features.Post processing is
Problems of fingerprint recognition
then done involving removal of spurious
Poor quality fingerprint images might
minutia patterns. result in lower accuracy. Factors such
as finger injuries or manual working,
can result in certain users being unable
to use a fingerprint-based recogni-
tion system, either temporarily or
permanently.Small-area sensors embed-
Matching ded in portable devices may result in
Finally matching of existing sample less information .skin conditions (e.g.,
and live sample is done with various dryness, wetness, dirtiness, temporary
matching techniques. or permanent cuts and bruises), sensor

3
conditions (e.g, dirtiness, noise, size) etc.
Some of these factors cannot be avoided
and some of them vary along time.
Poor quality images result in spurious
and missed features, thus degrading the
performance of the overall system.
Cost estimation
Secugen,ESSl Fingerprint scanners costs
around 3000 rupees for an optical based
system.
Capacitive based device cost little more
and most of the latest smartphones
come with these scanners embedded in
them.
FACE RECOGNITION SYS-
TEM
Facial images are probably the most
common biometric characteristic used by
humans to make a personal recognition.
The applications of facial recognition
range from a static, controlled verifi-
cation to a dynamic, uncontrolled face
identification in a cluttered background
(e.g., airport). The most popular ap-
proaches to face recognition are based
on either
1.The location and shape of facial
attributes, such as the eyes, eyebrows,
nose, lips, and chin and their spatial
relationships
2.The overall (global) analysis of the
face image that represents a face as a
weighted combination of a number of
canonical faces.
While the verification performance of
the face recognition[4] systems that are
commercially available is reasonable,
they impose a number of restrictions
on how the facial images are obtained,
sometimes requiring a fixed and simple
background or special illumination.
These systems also have difficulty in
recognizing a face from images captured
from two drastically different views and
under different illumination conditions.
It is questionable whether the face itself,
without any contextual information, is a
sufficient basis for recognizing a person
from a large number of identities with Face Recognition has been researched
4
an extremely high level of confidence.
In order that a facial recognition sys-
tem works well in practice, it should
automatically detect whether a face is
present in the acquired image,locate the
face if there is one, recognize the face
from a general viewpoint (i.e., from any
pose).
Hardware
With regards to hardware it depends
on how we are deploying the system. If
we are building an internet connected
product - with an embedded camera -
we would be sending up the images to
a cloud API. No processing would be
happening on device, apart from the
capture of images/video.
For local face recognition (offline and/or
on machine), it is recommended to
have mini-PC configurations due to
their small form factor and the minimal
hardware requirements face analysis
systems need. CPU power, and not
camera quality, plays the largest role
in detecting a large number of faces
simultaneously. In addition 1 GB of ram
or higher is sufficient. Embedding an
SDK in an app is a powerful and flexible
way of using face recognition without
the need for an internet connection.
The most common approach to define a
face recognition system is classified as
a two step process which are Face De-
tection and Feature Extraction Feature
Recognition.
The following is an outline of a face
recognition system
for several years, different type of meth-
ods tried to solve two major problems
in this area face identification , and face
verification.
Face identification try to indicate the
identity of a visitor. Fisherfaces and
Local Binary Patterns (LBP) perform
well in this function; In the other hands,
face verification, which try to verify if a
visitor is truly as his claim. Facial Trait
Code (FTC) and the Ensemble Voting
Algorithm (EVA) perform well.
Different Techniques Used for Face
Recognition
Principle component analysis (PCA)[5]
is widely used for dimensionality re-
duction, Turk and Pentland introduced
Eigenfaces in 1991. By this method,
the dimensionality of a face model can
be reduced from image pixel size to
several principle basis, the basis may
encode sufficient information about
the face. However, it is designed in
a way to best preserve data in the
embedding space, and consequently
cannot promise good discriminating
capability. Linear discriminant analysis
(LDA) is also used for dimensional-
ity reduction, and it provide a good
discriminating capability. Fisherfaces
improve the face recognition system,
but the drawback of this method is it
can not perform well in face verification.
LBP is a powerful method to solve face
recognition problem, it is efficient and
also easy to implement. The drawback
is it is hard to determine a verification
threshold, which mean the boundary
between imposter and guest is hard to
determine in chi square distance matrix
space. FTC has good performance
on verification, its distance matrix in
discrete integer space helping us tuning
verification easier than other methods.
In contrast to setting the threshold,
tuning the parameters of FTC are really
complicated and time-consuming, we do
lots of efforts on this algorithm.
Ensemble Voting Algorithm (EVA)
tries to gather different opinions of face
5
recognition algorithms, and votes for a
common idea for final decision. One of
the report when went through where
the algorithm was tested it achieved 95
percent accuracy in a random sample 20
faces set contained imposter and gallery
face.
Problems faced during Face Recog-
nition
Following are some of the problems
faced during face recognition.
1. Variations in skin color under differ-
ent lighting conditions
2. Presence of eyeglasses, make up,
pimples
3. Different face angles with respect to
the camera
4. Clarity of the face image, distance
from the camera
5. Different facial expressions/ Poses
(e.g., varying levels of smile)
6. Hats, caps, or hairstyles that may
partially cover the face
To reduce pose variation and illumina-
tion in extracted faces two extra actions
can be performed in pre-processing stage
to improve recognition results: 1. Eyes
detection can be used to remove head
turn, tilt, slant and position of face
2. Histogram equalization can also be
performed
Cost Estimation
Since the hardware requirements are
very much affordable ( all you need is
a camera and a board to process the
images ), face recognition system has a
very low cost estimate when compared
to other biometrics.
IRIS RECOGNITION SYSTEM
Biometric iris recognition[6] utilizes
pattern recognition systems based on
high-resolution and distortion-free im-
ages of the irises of the human eyes.Iris
is an organ whose structure remains
stable throughout life. Thus it serves as
a very good biometric for establishing
identity of an individual.
Iris recognition is considered to be the
most reliable and accurate biometric
identification system available. Iris
recognition system captures an image
of an individuals eye, the iris in the
image is then meant for the further
segmentation and normalization for
extracting its feature. The performance
of iris recognition systems depends on
the process of segmentation.
Hardware
To obtain an image of the Iris, cameras
are used; initial applications worked
in the visible range, acquiring colour
images of the users iris. These cameras
presented a major disadvantage which
had to be solved: lighting. The ambient
light of the room where the camera is
located may be relatively low thus a
flash is required and it is reflected in the
users iris. These reflections influence iris
measurements, the part of the iris where
this reflection is located is discarded,
also the system is not comfortable
for users due to the flash. For these
reasons, most iris cameras working in
the infrared range are now used. The
infrared region provides multiple ad-
vantages when compared to the visible
range: iris ridges are more defined; the
border between the iris and the pupil
is more pronounced; and the user is
not subjected to exposure of annoying
flashes. Steps in iris recognition[7]
1) Image Acquisition
2) Pattern Extraction
3) Matching
Image Acquisition
It deals with capturing of a high quality
image of the iris. Concerns on the
image acquisition rigs, Obtain images
with sufficient resolution and sharpness.
Good contrast in the iris pattern with
6
proper illumination. Well centered with-
out unduly constraining the operator.
Distance up to 3 meter. Near-infrared
camera or LED
Image preprocessing
It is generally done before Pattern
extraction which uses different methods
like Histogram Equalization to increase
Image contrast and Binarization .
Pattern Extraction
The system is to be composed of a num-
ber of subsystems, which correspond
to each stage of iris recognition. These
stages are segmentation locating the
iris region in an eye image, removing
the eyelids,eyelashes preset in the iris
region , normalisation creating a di-
mensionally consistent representation
of the iris region, and feature encoding
creating a template containing only
the most discriminating features of the
iris. The input to the system will be
an eye image, and the output will be
an iris template, which will provide a
mathematical representation of the iris
region.
Daugmans Algorithm is used in Seg-
mentation phase
To get a bit-wise biometric template
convolve the normalized iris region with
1D Log-Gabor filters.
Daugmans Rubber sheet model is used
for Normalization.
Matching
Hamming distance
In comparing the bit patterns X and Y,
the Hamming distance, HD, is defined
as the sum of disagreeing bits (sum of
the exclusive-OR between X and Y)
over N, the total number of bits in the
bit pattern. If two patterns are derived
from same iris, the hamming distance
between them will be close to 0 due to
high correlation.
Cost Estimation
Iritechs Iris Scanner costs about 9k
whereas other Iris Scanner costs about
25k.The major disadvantage is it is
expensive. Iris scanners are about thrice
the cost of fingerprint scanners.
Efficiency
Segmentation is the most critical stage,
since areas that are wrongly identified
as iris regions will corrupt biometric
templates resulting in poor recognition.
When tested on CASIA Database gave
FAR of 0.005% and FRR of 0.23%.
Disadvantages of Iris Recognition
One of the disadvantages with iris
recognition technology, being that it is
a relatively new technology, is that the
equipment is very expensive. The iris
is a small target and a scan cannot be
performed properly if the person is more
than a few meters away.
From the above study, the conclusion
can be tabulated[8] as follows:
4 Work Done
Got familiar with the existing Atten-
dance Management System using bio-
metrics(fingerprint) which is currently
being used in our department.
5 Future Work
1.Detailed study on Multimodal biomet-
ric systems and finger vein authentica-
tion.Hardware and software design for
Finger Vein.
2.Designing and Implementing a
database and a standalone software
solution which is independent of the
software provided by the vendor.
3.Finalizing the detailed design of the
solution.
References
[1] Anil K Jain, Jianjiang Feng, and
Karthik Nandakumar. Finger-
print matching. In: Computer
43.2 (2010).
[2] Hong Zhang and Xinsheng Wang.
A new fingerprint enhancement
algorithm. In: Broadband Net-
work and Multimedia Technology
(IC-BNMT), 2010 3rd IEEE In-
ternational Conference on. IEEE.
2010, pp. 12221225.
[3] Muzhir Shaban Al-Ani. A novel
thinning algorithm for finger-
print recognition. In: Interna-
tional Journal of Engineering
Sciences 2.2 (2013), pp. 4348.
[4] Wenyi Zhao et al. Face recog-
nition: A literature survey. In:
ACM computing surveys (CSUR)
35.4 (2003), pp. 399458.
[5] Timo Ahonen, Abdenour Hadid,
and Matti Pietikainen. Face de-
scription with local binary pat-
terns: Application to face recog-
nition. In: IEEE transactions on
pattern analysis and machine in-
telligence 28.12 (2006), pp. 2037
2041.
[6] Libor Masek et al. Recognition
of human iris patterns for biomet-
ric identification. In: (2003).
[7] Prateek Verma et al. Daugh-
mans algorithm method for iris
recognition - a biometric ap-
proach. In: International Jour-
nal of Emerging Technology and
Advanced Engineering 2.6 (2012),
pp. 177185.
[8] Rupinder Saini and Narinder
Rana. Comparison of various
biometric methods. In: Interna-
tional Journal of Advances in Sci-
ence and Technology (IJAST) Vol
2 (2014).