Face recognition using artificial neural network

PRESENTED BY:- GUIDED BY:-

SUMEET S. KAKANI PROF. V. M. UMALE

1

 Introduction
 What is an Artificial Neural Network?
 ANN Structure
 System Overview
 Local image sampling
 The Self Organizing-Map
 Perceptron learning rule
 Convolutional Network
 System details
 Implementation
 Applications
 Conclusion
 References

2

1. Why face recognition?
 Face Recognition systems enhance security, provide secure
access control, and protect personal privacy
 Improvement in the performance and reliability of face
recognition
 No need to remember any passwords or carry any ID

2. Why neural network?
 Adaptive learning: An ability to learn how to do tasks
 Self-Organization: An ANN can create its own organisation
 Remarkable ability to derive meaning from complicated or
imprecise data

3

 Definition: A Computing system made of a number of simple,
highly interconnected processing elements, which process
information by their dynamic state response to external input.

 Motivated right from its inception by the recognition.

 A machine that is designed to model the way in which the brain
performs a particular task.

 A massively parallel distributed processor.

 Resembles the brain in two respects:

1. Knowledge is acquired through a learning process.

2. Synaptic weights, are used to store the acquired knowledge.

4

 A Layer with n inputs xi.

 correspondent weights Wji (i=1, 2, 3...n)

 The function ∑ sums the n weighted inputs and passes
the result through a non-linear function φ(•), called
the activation function.

 The function φ (•) processes the adding results plus a
threshold value θ, thus producing the output Y.

5

a) Processing units
 Receive input
 The adjustment of the weights
 Three types of units: 1. input units 2. hidden units 3. output units
 During operation, units can be updated either synchronously or
asynchronously.

b) Connections between units
n
Y(t) = φ( ∑ wij(t) Xj(t) + θ)
j=1
c) Transfer Function:
The behavior of an ANN depends on both the weights and the
transfer function that is specified for the units.
This function typically falls into one of two categories:

 Linear : the output activity is proportional to the total weighted
output.

 Threshold: the output is set at one of two levels, depending on
whether the total input is greater than or less than some threshold
value. 6

Following are the basic processes that are used by the system
to capture and compare images:

1. Detection:
Recognition software searches the field of view of a video
camera for faces.
Once the face is in view, it is detected within a fraction of a
second .

2. Alignment:
Once a face is detected, the head's position, size and pose
is the first thing that is determined.

3. Normalization:
The image of the head is scaled and rotated so that it can be
registered and mapped into an appropriate size and pose.

7

4. Representation:
Translation of facial data into unique code is done
by the system.

5. Matching:
The newly acquired facial data is compared to the
stored data and (ideally) linked to at least one
stored facial representation
This includes-
Local image sampling
The Self Organizing-Map
Convolutional Network

8

Figure : A representation of the local image sampling process.

• We have evaluated two different methods of representing
local image samples.

• In each method a window is scanned over the image as
shown in figure.

9

 Used to reduce the dimensions of the image vector

 Self-Organizing Map(SOM), is an unsupervised learning
process, which learns the distribution of a set of patterns
without any class information.

 Unsupervised learning:
a) No external teacher and is based upon only local information.
b) It is also referred to as self-organization unsupervised learning.

The basic SOM consists of:
I) A 2 dimensional lattice L of neurons.
II) Each neuron ni belongs to L has an associated codebook
vector μi belongs to Rn.
III) The lattice is either rectangular or hexagonal as shown in figure

10

Figure: Rectangular and hexagonal lattice

• Self-organizing maps learn both the distribution and topology of the
input vectors they are trained on.

• Here a self-organizing feature map network identifies a winning
neuron i 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

11

 Perceptrons are trained on examples of
desired behavior.

 The desired behavior can be summarized by a set of input,
output pairs as:
p1t1, p2t2,………….. pQtQ

 Where p is an input to the network and t is the corresponding
correct (target) output.

 The perceptron learning rule can be written more succinctly in
terms of the error e = t - a, and the change to be made to the
weight vector Δw:

Case 1: If e = 0, then make a change Δw equal to 0.

Case 2: If e = 1, then make a change Δw equal to p T.

Case 3: If e = -1, then make a change Δw equal to -p T.
12

Figure : A typical convolutional network.

13

 Automatically synthesize simple problem specific
feature extractor from training data.

 Feature detectors applied everywhere.

 Features get progressively more global and invariant.

 The whole system is trained “end-to-end” with
gradient based method to minimize a global loss
function.

 Integrate segmentation, feature extraction, and
invariant classification in one stretch.

14

1. Convolutional Mechanism
Capable of extracting similar features in different
places in the image.
Shifting the input only shifts the feature map
(robust to shift).

2. Subsampling Mechanism
The exact positions of the extracted features
are not important.
Only relative position of a feature to another
feature is relevant.
Reduce spatial resolution – Reduce sensitivity
to shift and distortion.
15

 For the images in the training set, a fixed size window is
stepped over the entire image.
 As shown in earlier figure and local image samples are
extracted at each step.
 A self-organizing map is trained on the vectors from the
previous stage
16

 The same window as in the first step is stepped over all
of the images in the training and test sets.

 The local image samples are passed through the SOM
at each step, thereby creating new training and test
sets in the output space created by the self-organizing
map.

 A convolutional neural network is trained on the
newly created training set

17

Training: Testing:

18

 Law Enforcement: Minimizing victim trauma by narrowing mug
shot searches, Verifying identify for court records, and
comparing school Surveillance camera Images to known child
molesters.

 Security/Counter terrorism: Access control, comparing
surveillance images to Known terrorists.

 Day Care: Verify identity of individuals picking up the children.

 Missing Children/Runaways Search surveillance images and the
internet
for missing children and runaways.

 Residential Security: Alert homeowners of approaching
personnel.
 Healthcare: Minimize fraud by verifying identity.
 Banking: Minimize fraud by verifying identity.

19

 There are no explicit three-dimensional models
in our system, however we have found that the
quantized local image samples used as input
to the convolutional network represent
smoothly changing shading patterns.
 Higher level features are constructed from
these building blocks in successive layers of the
convolutional network.
 In comparison with the eigenfaces approach,
we believe that the system presented here is
able to learn more appropriate features in
order to provide improved generalization.

20

1. Steve Lawrence, C. Lee Giles , “Face Recognition: A Convolutional Neural Network
Approach”, IEEE transaction, St. Lucia, Australia.

2. David a brown, Ian craw, Julian lewthwaite, “Interactive Face retrieval using self
organizing maps-A SOM based approach to skin detection with application in real
time systems”, IEEE 2008 conference, Berlin, Germany.

3. Shahrin Azuan Nazeer, Nazaruddin Omar' and Marzuki Khalid, “Face Recognition
System using Artificial Neural Networks Approach”, IEEE - ICSCN 2007, MIT Campus,
Anna University, Chennai, India. Feb. 22-24, 2007. pp.420-425.

4. M. Prakash and M. Narasimha Murty, “Recognition Methods and Their Neural-
Network Models”, IEEE TRANSACTIONS ON NEURAL NETWORKS, VOL. 8, NO. 1, JANUARY
2005.

5. E. Oja, “Neural networks, principal components and subspaces”, Int. J. Neural Syst., vol.
1, pp. 61–68, 2004.

21

Face recognition using artificial neural network

More Related Content

What's hot (20)

Similar to Face recognition using artificial neural network (20)

Recently uploaded (20)

Face recognition using artificial neural network