1. ATM TRANSACTION USING FACE
RECOGNITON
PROJECT REPORT
Submitted in partial fulfillment of the requirements for the award of the Degree of Bachelor of Technology in
Electronics and Communication Engineering of the University of Kerala.
By
AJIN SUDHIR (11407001)
EJAZ NAVAS. S (11407007)
JAISON JOSEPH (11407008)
ROBIN GEORGE (11407017)
(Eighth Semester)
DEPARTMENT OF ELECTRONICS AND COMMUNICATION ENGINEERING
MARIAN ENGINEERING COLLEGE
KAZHAKUTTOM, TRIVANDRUM
APRIL 2015
2. DEPARTMENT OF ELECTRONICS & COMMUNICATION,
MARIAN ENGINEERING COLLEGE, KAZHAKUTTOM
CERTIFICATE
This is to certify that project work entitled “ATM TRANSACTION USING FACE
RECOGNITION” is bonafide record of project work carried out by AJIN SUDHIR, EJAZ
NAVAS. S, JASION JOSEPH and ROBIN GEORGE during the year 2015 in partial
fulfillment of the requirement for the award of the Degree of Bachelor of Technology in
Electronics & Communication Engineering of the University of Kerala, through MARIAN
ENGINEERING COLLEGE, Thiruvananthapuram under the guidance of Ms.Mena Raman.
Project coordinator Project Guide Head of Department
Internal Examiner External Examiner
3. ACKNOWLEDGEMENT
It is indeed a pleasure and a moment of satisfaction for expressing our gratitude and sincere
thanks to our project guide Ms.Mena Raman, Assistant Professor and our project coordinators,
Prof Dr.R.S Moni and Mrs. Ramola Joy, Assistant Professor of Electronics and
Communication Department who have been a constant source of encouragement.
We take this opportunity to express gratitude to our Principal Prof.Tomy Michael and the
management of Marian Engineering College for giving us the opportunity to express our ideas.
We are deeply indebted to Prof Dr.M. Sasi Kumar, Head of the Department, Electronics and
Communication.
We express our thanks to our parents and friends for extending their help towards the successful
completion of our project.
Finally, we express our heartfelt veneration to GOD almighty that helped us throughout this
endeavor.
4. ABSTRACT
Automatic Teller Machines (ATMs) are widely used in our daily lives due to their convenience,
wide-spread availability and time-independent operation. Automatic retraction of forgotten card
or cash by ATMs is a problem with serious consequences (lost time and money), typically
caused by user inattention/negligence. In this work, we propose a scheme in which the retraction
rate of an ATM is decreased using face detection and recognition methods via ATM's built-in
camera. There is an urgent need for improving security in banking region. With the birth of the
Automatic Teller Machines, banking became a lot easier though with its own troubles of
insecurity. Due to tremendous increase in the number of criminals and their activities, the ATM
has become insecure. ATM systems today use no more than an access card and PIN for identity
verification. The recent progress in biometric identification techniques, including finger printing,
retina scanning, and facial recognition has made a great efforts to rescue the unsafe situation at
the ATM. This research looked into the development of a system that integrates facial
recognition technology into the identity verification process used in ATMs. An ATM model that
is more reliable in providing security by using facial recognition software is proposed .The
development of such a system would serve to protect consumers and financial institutions alike
from intruders and identity thieves. This project proposes an automatic teller machine security
model that would combine a physical access card, a PIN, and electronic facial recognition that
will go as far as withholding the fraudster‟s card. If this technology becomes widely used, faces
would be protected as well as PINs. However, it obvious that man‟s biometric features cannot be
replicated, this proposal will go a long way to solve the problem of account safety making it
possible for the actual account owner alone have access to his accounts. The combined biometric
features approach is to serve the purpose both the identification and authentication that card and
PIN do.
6. LIST OF IMAGES
Image No. Image Name Page No.
1.1
A Person Inserting An ATM
Card In An ATM Machine 01
2.1
Theft Survey: 75% - ATM
card related thefts, 18% -
Multiple ATM machines,
7% - Robbery
02
2.2 Newspaper Clipping of ATM
fraud
03
3.1
Points where facial
recognition technology
extracts the features for
identification
07
4.1 Basic Layout Of Proposed
System
09
4.2 Block Diagram Of Proposed
System
10
4.3 Circuit Diagram 13
5.1 Processes involved 14
7. 5.2 Original Image and
Compressed Image using DCT
17
5.3 Capturing The Image of User 24
5.4 Training the Databases 24
5.5 After Training the Network 25
5.6 Database Created 25
5.7 Face Recognized from Library 26
5.8 Product 26
5.9 Various Stages 27
6.1 Solder layer and Component
layer
28
8. 1. INTRODUCTION
For the past several decades‟ easy access, secure storage of money was a problem for humans.
Then came a solution to that problem. Banks provided an ATM card with which we can
withdraw money from our account. Each person was given a set of cards, but those cards were
only for one time use. Each time a person withdraws money from his/her account, the card is
punched making it unusable. So the introduction and concept of those cards was did not last long
because, each time the card runs out they have to go to the bank requesting a set of new cards.
Then came the modern ATM cards, which have a magnetic strip on the back side of it. Each
person‟s account will have a unique ATM card which has a unique set of codes behind the
magnetic strip. The codes are ciphered. When inserted into an ATM (Automated Teller Machine)
machine, a scanner scans the ciphered codes behind the magnetic strip. It then asks for a four
digit PIN code which is kind of a password for protecting our account. It matches the PIN with
the database and if the codes are matched then, we will be able to access our account. But if the
PIN number does not match them self then we won‟t be able to access our account. This is the
current technology which is present regarding ATM transactions.
But there are some shortcomings for this technology. Elder people naturally tend to forget their
PIN number, so they usually write it down on their ATM card‟s cover. If the card gets lost and if
fallen into the wrong hands then, all their hard earned saving will be lost in matter of seconds.
Fig1.1: A Person Inserting An ATM Card In An ATM Machine
9. 2. LITERATURE SURVEY
In a survey conducted by the Government of India, it was very evident that thefts occur
mainly related to ATM cards.
Fig2.1: Theft Survey: 75% - ATM card related thefts, 18% - Multiple ATM machines,
7% - Robbery
Thefts related to ATM cards are the main problem. Duplication of the codes encrypted
behind the magnetic strips can be replicated. India‟s first ATM fraud case was by duplicating
the magnetic strip and the codes ciphered behind it. He looted many peoples bank accounts,
and went to jail for robbery. He got help from a website which charges a handsome fee in
exchange for the ciphered codes and PIN numbers through which they access other people‟s
accounts and steal their hard earned money from their account without their notice.
10. Fig2.2: Newspaper Clipping of ATM fraud
Multiple ATM machines also provide the perfect platform for looting other peoples accounts.
In Navi Mumbai an incident took place in an ATM counter where there were 4 ATM
machines. When the victim tried in the first ATM money transaction did not take place he
went to the second machine and even then transaction did not take place, a person standing
there told him that the fourth machine was working, he went there and got the cash. After one
hour a message was sent to his mobile that Rs.50000 was deducted from his account. He
immediately reported this to the police and the bank authorities. The bank immediately
refunded his lost amount due to their lack of security measures. Like this many incidents are
happening all over the country, with more sophisticated techniques.
11. 3. FACE RECOGNITON
Facial recognition (or face recognition) is a type of biometric software application that can
identify a specific individual in a digital image by analyzing and comparing patterns.
Facial recognition systems are commonly used for security purposes but are increasingly being
used in a variety of other applications. The Kinect motion gaming system, for example, uses
facial recognition to differentiate among players.
Most current facial recognition systems work with numeric codes called faceprints. Such systems
identify 80 nodal points on a human face. In this context, nodal points are end points used to
measure variables of a person‟s face, such as the length or width of the nose, the depth of the eye
sockets and the shape of the cheekbones. These systems work by capturing data for nodal points
on a digital image of an individual‟s face and storing the resulting data as a faceprint. The
faceprint can then be used as a basis for comparison with data captured from faces in an image or
video.
Facial recognition systems based on faceprints can quickly and accurately identify target
individuals when the conditions are favorable. However, if the subject‟s face is partially
obscured or in profile rather than facing forward, or if the light is insufficient, the software is less
reliable. Nevertheless, the technology is evolving quickly and there are several emerging
approaches, such as 3D modeling, that may overcome current problems with the systems.
According to the National Institute of Standards and Technology (NIST), the incidence of false
positives in facial recognition systems has been halved every two years since 1993 and, as of the
end of 2011, was just .003%
Currently, a lot of facial recognition development is focused on smartphone applications.
Smartphone facial recognition capacities include image tagging and other social networking
integration purposes as well as personalized marketing. A research team at Carnegie Mellon has
developed a proof-of-concept iPhone app that can take a picture of an individual and -- within
seconds -- return the individual's name, date of birth and social security number.
12. Facial recognition system software is a computer aided application used mainly for
automatically identifying or verifying a person from a digital image or a video frame from
a video source. One of the ways to do this is by comparing selected facial features from the
image and a facial database. Techniques:
3.1 Traditional
Some facial recognition algorithms identify facial features by extracting landmarks, or features,
from an image of the subject's face. For example, an algorithm may analyze the relative position,
size, and/or shape of the eyes, nose, cheekbones, and jaw. These features are then used to search
for other images with matching features. Other algorithms normalize a gallery of face images
and then compress the face data, only saving the data in the image that is useful for face
recognition. A probe image is then compared with the face data. One of the earliest successful
systems is based on template matching techniques applied to a set of salient facial features,
providing a sort of compressed face representation.
Recognition algorithms can be divided into two main approaches, geometric, which looks at
distinguishing features, or photometric, which is a statistical approach that distills an image into
values and compares the values with templates to eliminate variances.
Popular recognition algorithms include Principal Component Analysis using Eigen faces, Linear
Discriminate Analysis, Elastic Bunch Graph Matching using the Fisher face algorithm,
the Hidden Markov model, the Multi linear Subspace Learning using tensor representation, and
the neuronal motivated dynamic link matching.
3.2 3-Dimensional recognition
A newly emerging trend, claimed to achieve improved accuracies, is three-dimensional face
recognition. This technique uses 3D sensors to capture information about the shape of a face.
This information is then used to identify distinctive features on the surface of a face, such as the
contour of the eye sockets, nose, and chin.
One advantage of 3D facial recognition is that it is not affected by changes in lighting like other
techniques. It can also identify a face from a range of viewing angles, including a profile
view. Three-dimensional data points from a face vastly improve the precision of facial
recognition. 3D research is enhanced by the development of sophisticated sensors that do a better
13. job of capturing 3D face imagery. The sensors work by projecting structured light onto the face.
Up to a dozen or more of these image sensors can be placed on the same CMOS chip—each
sensor captures a different part of the spectrum.
Even a perfect 3D matching technique could be sensitive to expressions. For that goal a group at
the Teknion applied tools from metric geometry to treat expressions as isometrics. A company
called Vision Access created a firm solution for 3D facial recognition. The company was later
acquired by the biometric access company Bios crypt Inc. which developed a version known as
3D Fast Pass.
3.3 Skin texture analysis
Another emerging trend uses the visual details of the skin, as captured in standard digital or
scanned images. This technique, called skin texture analysis, turns the unique lines, patterns, and
spots apparent in a person‟s skin into a mathematical space.
Tests have shown that with the addition of skin texture analysis, performance in recognizing
faces can increase 20 to 25 percent.
There are mainly three steps for this technology:
Recognition in this technology plays a major role, recognition used in the description of
biometric systems like facial recognition, finger print or iris recognition relating to their
fundamental function.
Verification is the task where the biometric system attempts to confirm an individual‟s
claimed identity by comparing a submitted sample to one or more previously enrolled
templates
Identification is the task where the biometric system searches a database for a reference
finding a match for the submitted biometric sample; a biometric sample is collected and
compared to all the templates in the database.
14. 3.4 Process involved
Sensor: A sensor collects all the data‟s and converts the information into a digital format.
Signal processing algorithms: This is where quality control activities and development of
the template takes place.
Data Storage: Keeps information that new biometric templates will be compared to the
photograph in the database.
Matching algorithm: Compares the new template to other templates in the data storage
Decision process: Uses the results from the matching component to make a system level
decision.
Fig3.1: Points where facial recognition technology extracts the features for identification.
15. 3.5 Advantages of Facial Recognition Technology
Eradicate Fraud costs for the bank.
Deliver a practical and workable solution that addresses the requirements of the
regulatory authorities.
Limit the financial risks given that they were forced to take responsibility for
financial loss [rather than being allowed to pass this on to the account-holder].
Provide a framework that still allowed for high withdrawal limits to cater for the
demands of a cash-focused customer base.
Take societal responsibility to reduce rising levels of crime that were associated
with cash-card transactions.
Increase customer satisfaction.
For the account-holder, the potential advantages are:
Different charges for transactions given that the transaction takes place in a more
secure manner.
Higher withdrawal and transaction limits.
Peace of mind given the higher level of security applied to the account.
3.6 ATM Fraud Types To Be Prevented By ATM Facial Recognition Technology
Unauthorized financial operations using lost or stolen cards and pin codes
which many inexperienced card owners write down on a card or sore the PIN
code together with the card.
Fraud based on Trust- The card or its duplicate can be used by a fraudster
without the permission of the card owner.
16. 4. PROPOSED SYSTEM
Every person who has an ATM card should go to the bank and take photographs of their face (at
least five) to create a library and store it in the database. Now when a user enters an ATM
counter, a camera will capture the users face and the face recognition algorithm is started for
identifying the user. When a match is found from the per defined data base, the user is said to be
recognized. When the face recognition is complete, a randomly generated PIN is sent to our
registered mobile phone as a second step security. When the user enters the randomly received
PIN he/she can access his/her account and withdraw cash. If the face cannot be matched, the
system captures the image of the user again. If the PIN is not entered within 10 seconds, the
system will give you an option to resend the code to your mobile, in case if code does not receive
to the users mobile.
4.1Algorithm
Take customer‟s picture(s) when account is opened
Take user‟s picture, attempt to match it to database image(s)
If match is successful, allow transaction
If match is unsuccessful, limit available transactions
Fig4.1: Basic Layout Of Proposed System
Image,
random PIN
User verified
Message
Image, account no,
PIN
User verified
Message
17. Fig4.2: Block Diagram of Proposed System
The RS232 provides an interconnection between PC and the micro controller; it acts as a level
translator. An RS-232 serial port was once a standard feature of a personal computer, used for
connections to modems, printers, mice, data storage, uninterruptible power supplies, and other
peripheral devices. In RS-232, user data is sent as a time-series of bits. Both synchronous and
asynchronous transmissions are supported by the standard. In addition to the data circuits, the
standard defines a number of control circuits used to manage the connection between the DTE
and DCE. Each data or control circuit only operates in one direction, that is, signaling from a
DTE to the attached DCE or the reverse. Since transmit data and receive data are separate
circuits, the interface can operate in a full duplex manner, supporting concurrent data flow in
18. both directions. The standard does not define character framing within the data stream, or
character encoding.However, RS-232 is hampered by low transmission speed, large voltage
swing, and large standard connectors. In modern personal computers, USB has displaced RS-232
from most of its peripheral interface roles. Many computers do not come equipped with RS-232
ports and must use either an external USB-to-RS-232 converter or an internal expansion card
with one or more serial ports to connect to RS-232 peripherals. RS-232 devices are widely used,
especially in industrial machines, networking equipment and scientific instruments. An RS-
232 serial port is a standard feature of personal computer, used for inter connections
to modems, printers, mice, data storage, uninterruptible power supplies, and other peripheral
devices. However, RS-232 is hampered by low transmission speed, large voltage swing, and
large standard connectors. In modern personal computers, USB has displaced RS-232 from most
of its peripheral interface roles. Many computers do not come equipped with RS-232 ports and
must use either an external USB-to-RS-232 converter or an internal expansion card with one or
more serial ports to connect to RS-232 peripherals. RS-232 devices are widely used, especially in
industrial machines, networking equipment and scientific instruments.
GSM is a cellular network, which means that cell phones connect to it by searching for cells in
the immediate vicinity. There are five different cell sizes in a GSM network—
macro, micro, pico, femto, and umbrella cells. The coverage area of each cell varies according to
the implementation environment. Macro cells can be regarded as cells where the base
station antenna is installed on a mast or a building above average rooftop level. Micro cells are
cells whose antenna height is under average rooftop level; they are typically used in urban areas.
Picocells are small cells whose coverage diameter is a few dozen metres; they are mainly used
indoors. Femtocells are cells designed for use in residential or small business environments and
connect to the service provider‟s network via a broadband internet connection. Umbrella cells are
used to cover shadowed regions of smaller cells and fill in gaps in coverage between those cells.
Cell horizontal radius varies depending on antenna height, antenna gain, and propagation
conditions from a couple of hundred meters to several tens of kilometres. The longest distance
the GSM specification supports in practical use is 35 kilometres (22 mi). There are also several
implementations of the concept of an extended cell where the cell radius could be double or even
more, depending on the antenna system, the type of terrain, and the timing advance.
19. Indoor coverage is also supported by GSM and may be achieved by using an indoor picocell base
station, or an indoor repeater with distributed indoor antennas fed through power splitters, to
deliver the radio signals from an antenna outdoors to the separate indoor distributed antenna
system. These are typically deployed when significant call capacity is needed indoors, like in
shopping centers or airports. GSM networks operate in a number of different carrier frequency
ranges (separated into GSM frequency ranges for 2G and UMTS frequency bands for 3G), with
most 2GGSM networks operating in the 900 MHz or 1800 MHz bands. Where these bands were
already allocated, the 850 MHz and 1900 MHz bands were used instead However, this is not a
prerequisite, since indoor coverage is also provided by in-building penetration of the radio
signals from any nearby cell.
The camera takes a photograph of the user and sends to the database for matching. A matching
algorithm is run by using an algorithm. Once the matching algorithm is run, and when it is
successful, a randomly generated number is sent to the user‟s mobile phone. For that provision a
GSM module is kept. GSM (Global System for Mobile Communications, originally Groupe
Spécial Mobile), is a standard developed by the European Telecommunications Standards
Institute(ETSI) to describe protocols for second generation (2G) digital cellular networks used
by mobile phones.
21. 5. PROCESSES INVOLVED
Read Image via webcam (512×512 pixels)
RGB to Gray scale conversion
Image Resizing (8×8 pixels)
2D-DCT of Image
Creation of SOM Neural Network
Training the Neural Network
Create Image Database
Fig5.1: Processes involved
The image is of a user is first captured using a 2.0 megapixel camera. The image captured will be
of dimension 512x512 in size. The captured image is then converted to grayscale image i.e to
black and white, since black and white images will have greater details and greater contrast than
the RGB image. It is then converted to an 8 x 8 image.
22. 5.1 IMAGE COMPRESSION
The discrete cosine transform is an algorithm widely used in different applications. The most
popular use of the DCT is for data compression, as it forms the basis for the international
standard loss image compression algorithm known as JPEG. A discrete cosine transform (DCT)
expresses a finite sequence of data points in terms of a sum of cosine functions oscillating at
different frequencies. DCTs are important to numerous applications in science and engineering,
from lossy compression of audio (e.g. MP3) and images (e.g. JPEG) (where small high-
frequency components can be discarded), to spectral methods for the numerical solution
of partial differential equations. The use of cosine rather than sine functions is critical for
compression, since it turns out (as described below) that fewer cosine functions are needed to
approximate a typical signal, whereas for differential equations the cosines express a particular
choice of boundary conditions. Like any Fourier-related transform, discrete cosine transforms
(DCTs) express a function or a signal in terms of a sum of sinusoids with different
frequencies and amplitudes. Like the discrete Fourier transforms (DFT), a DCT operates on a
function at a finite number of discrete data points. The obvious distinction between a DCT and a
DFT is that the former uses only cosine functions, while the latter uses both cosines and sines (in
the form of complex exponentials). However, this visible difference is merely a consequence of a
deeper distinction: a DCT implies differentboundary conditions than the DFT or other related
transforms. However, because DCTs operate on finite, discrete sequences, two issues arise that
do not apply for the continuous cosine transform. First, one has to specify whether the function is
even or odd at both the left and right boundaries of the domain (i.e. the min-n and max-
n boundaries in the definitions below, respectively). Second, one has to specify around what
point the function is even or odd. In particular, consider a sequence abcd of four equally spaced
data points, and say that we specify an even left boundary. There are two sensible possibilities:
either the data are even about the sample a, in which case the even extension is dcbabcd, or the
data are even about the point halfway between a and the previous point, in which case the even
extension is dcbaabcd (a is repeated)
In particular, a DCT is a Fourier-related transform similar to the discrete Fourier
transform (DFT), but using only real numbers. DCTs are equivalent to DFTs of roughly twice
23. the length, operating on real data with even symmetry (since the Fourier transform of a real and
even function is real and even), where in some variants the input and/or output data are shifted
by half a sample. There are eight standard DCT variants, of which four are common.
The most common variant of discrete cosine transform is the type-II DCT, which is often called
simply "the DCT" its inverse, the type-III DCT, is correspondingly often called simply "the
inverse DCT" or "the IDCT". Two related transforms are the discrete sine transforms (DST),
which is equivalent to a DFT of real and odd functions, and the modified discrete cosine
transforms (MDCT), which is based on a DCT of overlapping data.
The DCT has the property that, for a typical image, most of the visually significant information
about the image is concentrated in just a few coefficients. Extracted DCT coefficients can be
used as a type of signature that is useful for recognition tasks, such as face recognition. Face
images have high correlation and redundant information which causes computational burden in
terms of processing speed and memory utilization. The DCT transforms images from the spatial
domain to the frequency domain. Since lower frequencies are more visually significant in an
image than higher frequencies, the DCT discards high-frequency coefficients and quantizes the
remaining coefficients. This reduces data volume without sacrificing too much image quality.
The proposed technique uses the DCT transform matrix in the MATLAB Image Processing
Toolbox. This technique is efficient for small square inputs such as image blocks of 8 × 8 pixels.
The proposed design technique calculates the 2D-DCT of the image blocks of size 8 × 8 pixels
using „8‟ out of the 64 DCT coefficients for masking. The other 56 remaining coefficients are
discarded (set to zero). The image is then reconstructed by computing the 2D-IDCT of each
block using the DCT transform matrix computation method. Finally, the output is a set of arrays.
Each array is of size 8 × 8 pixels and represents a single image. Empirically, the upper left corner
of each 2D-DCT matrix contains the most important values, because they correspond to low-
frequency components within the processed image block.
24. Fig5.2: Original Image and Compressed Image Using DCT
5.2 SELF ORGANISING MAPS (SOM)
A self-organizing map (SOM) or self-organizing feature map (SOFM) is a type of artificial
neural network (ANN) that is trained using unsupervised learning to produce a low-dimensional
(typically two-dimensional), discretized representation of the input space of the training samples,
called a map. Self-organizing maps are different from other artificial neural networks in the
sense that they use a neighborhood function to preserve the topological properties of the input
space. The self-organizing map also known as a Kohonen Map is a well-known artificial neural
network. It is an unsupervised learning process, which learns the distribution of a set of patterns
without any class information. It has the property of topology preservation. The goal of learning
in the self-organizing map is to cause different parts of the network to respond similarly to
certain input patterns. This is partly motivated by how visual, auditory or
other sensory information is handled in separate parts of the cerebral cortex in the human brain.
The weights of the neurons are initialized either to small random values or sampled evenly from
the subspace spanned by the two largest principal component eigenvectors. With the latter
alternative, learning is much faster because the initial weights already give a good approximation
of SOM weights. The network must be fed a large number of example vectors that represent, as
25. close as possible, the kinds of vectors expected during mapping. The examples are usually
administered several times as iterations. This process is repeated for each input vector for a
(usually large) number of cycles λ. The network winds up associating output nodes with groups
or patterns in the input data set. If these patterns can be named, the names can be attached to the
associated nodes in the trained net.
During mapping, there will be one single winning neuron: the neuron whose weight vector lies
closest to the input vector. This can be simply determined by calculating the Euclidean distance
between input vector and weight vector. While representing input data as vectors has been
emphasized in this article, it should be noted that any kind of object which can be represented
digitally, which has an appropriate distance measure associated with it, and in which the
necessary operations for training are possible can be used to construct a self-organizing map.
This includes matrices, continuous functions or even other self-organizing maps.
There is a competition among the neurons to be activated or fired. The result is that only one
neuron that wins the competition is fired and is called the “winner”. A SOM network identifies a
winning neuron using the same procedure as employed by a competitive layer. However, instead
of updating only the winning neuron, all neurons within a certain neighborhood of the winning
neuron are updated using the Kohonen Rule. This makes SOMs useful for visualizing low-
dimensional views of high-dimensional data, akin to multidimensional scaling. The artificial
neural network introduced by the Finnish professor Teuvo Kohonen in the 1980s is sometimes
called a Kohonen map or network. The Kohonen net is a computationally convenient abstraction
building on work on biologically neural models from the 1970s and morphogenesis models
dating back to Alan Turing in the 1950s
Like most artificial neural networks, SOMs operate in two modes: training and mapping.
"Training" builds the map using input examples (a competitive process, also called vector
quantization), while "mapping" automatically classifies a new input vector.
A self-organizing map consists of components called nodes or neurons. Associated with each
node are a weight vector of the same dimension as the input data vectors, and a position in the
map space. The usual arrangement of nodes is a two-dimensional regular spacing in
a hexagonal or rectangular grid. The self-organizing map describes a mapping from a higher-
dimensional input space to a lower-dimensional map space. The procedure for placing a vector
26. from data space onto the map is to find the node with the closest (smallest distance metric)
weight vector to the data space vector.
While it is typical to consider this type of network structure as related to feed forward
networks where the nodes are visualized as being attached, this type of architecture is
fundamentally different in arrangement and motivation.
Useful extensions include using toroidal grids where opposite edges are connected and using
large numbers of nodes.
It has been shown that while self-organizing maps with a small number of nodes behave in a way
that is similar to K-means, larger self-organizing maps rearrange data in a way that is
fundamentally topological in character.
It is also common to use the U-Matrix. The U-Matrix value of a particular node is the average
distance between the node's weight vector and that of its closest neighbors. In a square grid, for
instance, we might consider the closest 4 or 8 nodes (the Von Neumann and Moore
neighborhoods, respectively), or six nodes in a hexagonal grid.
Large SOMs display emergent properties. In maps consisting of thousands of nodes, it is possible
to perform cluster operations on the map itself.
The Kohonen rule allows the weights of a neuron to learn an input vector, and because of this it
is useful in recognition applications. Hence, in this system, a SOM is employed to classify DCT-
based vectors into groups to identify if the subject in the input image is “present” or “not
present” in the image database. SOMs can be one-dimensional, two-dimensional or
multidimensional maps. The number of input connections in a SOM network depends on the
number of attributes to be used in the classification. Training images are mapped into a lower
dimension using the SOM network and the weight matrix of each image stored in the training
database. During recognition trained images are reconstructed using weight matrices and
recognition is through untrained test images using Euclidean distance as the similarity measure.
Training and testing for our system was performed using the MATLAB Neural Network
Toolbox. . Unsupervised Learning During the learning phase, the neuron with weights closest to
the input data vector is declared as the winner. Then weights of all of the neurons in the
27. neighborhood of the winning neuron are adjusted by an amount inversely proportional to the
Euclidean distance. It clusters and classifies the data set based on the set of attributes used. The
learning algorithm is summarized as follows:
Initialization: Choose random values for the initial weight vectors
Sampling: Draw a sample x from the input space with a certain probability.
Similarity Matching: Find the best matching (winning) neuron i(x) at time t, 0 < t ≤ n by
using the minimum distance Euclidean criterion
Updating: Adjust the synaptic weight vector of all neurons
Continue with step 2 until no noticeable changes in the feature map are observed.
During the training phase, labeled DCT-vectors are presented to the SOM one at a time. For each
node, the number of “wins” is recorded along with the label of the input sample. The weight
vectors for the nodes are updated as described in the learning phase. By the end of this stage,
each node of the SOM has two recorded values: the total number of winning times for subject
present in image database, and the total number of winning times for subject not present in image
database.
There are two ways to interpret a SOM. Because in the training phase weights of the whole
neighborhood are moved in the same direction, similar items tend to excite adjacent neurons.
Therefore, SOM forms a semantic map where similar samples are mapped close together and
dissimilar ones apart. This may be visualized by a U-Matrix (Euclidean distance between weight
vectors of neighboring cells) of the SOM.
The other way is to think of neuronal weights as pointers to the input space. They form a discrete
approximation of the distribution of training samples. More neurons point to regions with high
training sample concentration and fewer where the samples are scarce.
SOM may be considered a nonlinear generalization of Principal components analysis (PCA). It
has been shown, using both artificial and real geophysical data, that SOM has many
advantages over the conventional feature extraction methods such as Empirical Orthogonal
Functions (EOF) or PCA.
Originally, SOM was not formulated as a solution to an optimisation problem. Nevertheless,
there have been several attempts to modify the definition of SOM and to formulate an
28. optimisation problem which gives similar results. For example, Elastic maps use the mechanical
metaphor of elasticity to approximate principal manifolds the analogy is an elastic membrane
and plate
5.3 Training the neural network
There is no single formal definition of what an artificial neural network is. However, a class of
statistical models may commonly be called "Neural" if they possess the following characteristics:
1. consist of sets of adaptive weights, i.e. numerical parameters that are tuned by a
learning algorithm, and
2. are capable of approximating non-linear functions of their inputs.
The adaptive weights are conceptually connection strengths between neurons, which are
activated during training and prediction.
Neural networks are similar to biological neural networks in performing functions collectively
and in parallel by the units, rather than there being a clear delineation of subtasks to which
various units are assigned. The term "neural network" usually refers to models employed
in statistics, cognitive psychology and artificial intelligence. Neural network models which
emulate the central nervous system are part of theoretical neuroscience and computational
neuroscience.
In modern software implementations of artificial neural networks, the approach inspired by
biology has been largely abandoned for a more practical approach based on statistics and signal
processing. In some of these systems, neural networks or parts of neural networks (like artificial
neurons) form components in larger systems that combine both adaptive and non-adaptive
elements. While the more general approach of such systems is more suitable for real-world
problem solving, it has little to do with the traditional artificial intelligence connectionist models.
What they do have in common, however, is the principle of non-linear, distributed, parallel and
local processing and adaptation. Historically, the use of neural networks models marked a
paradigm shift in the late eighties from high-level (symbolic) AI, characterized by expert
systems with knowledge embodied in if-then rules, to low-level (sub-symbolic) machine
learning, characterized by knowledge embodied in the parameters of a dynamical system.
29. Once a network has been structured for a particular application, that network is ready to be
trained. To start this process the initial weights are chosen randomly. Then, the training, or
learning, begins.
There are two approaches to training - supervised and unsupervised. Supervised training involves
a mechanism of providing the network with the desired output either by manually "grading" the
network's performance or by providing the desired outputs with the inputs. Unsupervised training
is where the network has to make sense of the inputs without outside help.
The vast bulk of networks utilize supervised training. Unsupervised training is used to perform
some initial characterization on inputs.
In supervised training, both the inputs and the outputs are provided. The network then processes
the inputs and compares its resulting outputs against the desired outputs. Errors are then
propagated back through the system, causing the system to adjust the weights which control the
network. This process occurs over and over as the weights are continually tweaked. The set of
data which enables the training is called the "training set." During the training of a network the
same set of data is processed many times as the connection weights are ever refined.
The current commercial network development packages provide tools to monitor how well an
artificial neural network is converging on the ability to predict the right answer. These tools
allow the training process to go on for days, stopping only when the system reaches some
statistically desired point, or accuracy. However, some networks never learn. This could be
because the input data does not contain the specific information from which the desired output is
derived. Networks also don't converge if there is not enough data to enable complete learning.
Ideally, there should be enough data so that part of the data can be held back as a test. Many
layered networks with multiple nodes are capable of memorizing data. To monitor the network to
determine if the system is simply memorizing its data in some nonsignificant way, supervised
training needs to hold back a set of data to be used to test the system after it has undergone its
training. (Note: memorization is avoided by not having too many processing elements.)
If a network simply can't solve the problem, the designer then has to review the input and
outputs, the number of layers, the number of elements per layer, the connections between the
30. layers, the summation, transfer, and training functions, and even the initial weights themselves.
Those changes required to create a successful network constitute a process wherein the "art" of
neural networking occurs.
Another part of the designer's creativity governs the rules of training. There are many laws
(algorithms) used to implement the adaptive feedback required to adjust the weights during
training. The most common technique is backward-error propagation, more commonly known as
back-propagation. These various learning techniques are explored in greater depth later in this
report.
Yet, training is not just a technique. It involves a "feel," and conscious analysis, to insure that the
network is not over trained. Initially, an artificial neural network configures itself with the
general statistical trends of the data. Later, it continues to "learn" about other aspects of the data
which may be spurious from a general viewpoint.
When finally the system has been correctly trained, and no further learning is needed, the
weights can, if desired, be "frozen." In some systems this finalized network is then turned into
hardware so that it can be fast. Other systems don't lock themselves in but continue to learn
while in production use.
36. 7. CONCLUSION
We thus developed an ATM model that is more reliable in providing security by using facial
recognition software. By keeping the time elapsed in the verification process to a negligible
amount we even try to maintain the efficiency of this ATM system to a greater degree.
Biometrics as means of identifying and authenticating account owners at the Automated Teller
Machines gives the needed and much anticipated solution to the problem of illegal transactions.
In this project, we have tried to proffer a solution to the much dreaded issue of fraudulent
transactions through Automated Teller Machine by biometrics that can be made possible only
when the account holder is physically present. Thus, it eliminates cases of illegal transactions at
the ATM points without the knowledge of the authentic owner. Using a biometric feature for
identification is strong and it is further fortified when another is used at authentication level.
