You are on page 1of 20

1

WAES 3104:
Fundamentals of Artificial
Neural Network

Azri bin Mohd Khanil

WEK110006

Individual Assignment

TABLE OF CONTENTS

INTRODUCTION 3
SCOPE 5
lIMITATIONS & ASSUMPTIONS .. 6
CHARACTERSTICS... 7

SUITABLE ANN FOR FACE RECOGNITION SYSTEM.. 8

5.1 FACE RECOGNITION APPLICATION


5.1.1 INPUT.. 9
5.1.2 PROCESS.. 13
OUTPUT....15

6. CONCLUSION17

REFERENCES. 18

Introduction

The human face plays an important role in our social interaction, conveying
peoples identity. Using the human face as a key to security, biometric face
recognition technology has received significant attention in the past several
years due to its potential for a wide variety of applications in both law
enforcement and non-law enforcement.

As compared with other biometrics systems using fingerprint/palmprint and


iris, face recognition has distinct advantages because of its non-contact
process. Face images can be captured from a distance without touching the
person being identified, and the identification does not require interacting with
the person. In addition, face recognition serves the crime deterrent purpose
because face images that have been recorded and archived can later help
identify a person.

Automated facial recognition involves the identification of an individual based


on his or her facial geometry. For facial recognition to be successful, there
needs to be a quality digital image of an individuals face, a database of digital
images of identified individuals, and facial recognition software that will
accurately find a match between the two.

Of all biometric technologies, facial recognition most closely mimics how


people identify others: by scrutinizing their face. What is an effortless skill in
humans has proven immensely difficult and expensive to replicate in
machines. But through a convergence of factors in the past few years, facial
recognition has become increasingly accurate technology.

Digital images have become a trend, through the existence of surveillance


cameras, camera-equipped smart phones, and inexpensive high-quality digital
cameras. Cheap data storage has led to massive online databases of images
of identified individuals, such as licensed drivers, passport holders, employee
IDs and convicted criminals. Individuals have embraced online photo sharing
and photo tagging on platforms such as Facebook, Instagram, Picasa and
Flickr. There have also been significant improvements in facial recognition
technology, including advancements in analyzing images and extracting data.

Faces have been transformed into electronic information that can be analyzed
and categorized in unprecedented ways. What makes facial image data so
valuable, and so sensitive, is that it is a uniquely measurable characteristic of
our body and a key to our identity.

Some security applications of facial recognition technology are undoubtedly


beneficial, such as authentication of employees allowed to access nuclear
plant facilities, for example. Other example is in the 2000 presidential election,
the Mexican government employed facial recognition software to prevent
voter fraud. Some individuals had been registering to vote under several
different names, in an attempt to place multiple votes. By comparing new
facial images to those already in the voter database, authorities were able to
reduce duplicate registrations. At the same time, facial recognition holds
implications for privacy and for societal values in general.

2.1 Scope

With the expanded use of video surveillance cameras and camera


equipped mobile phones throughout the world, we are seeing increasing
rate of available crime scene video/still photo evidence that contains
suspect face images.
Law enforcement agencies can now use these crime scene latent face
images to identify possible persons of interest in these criminal
investigations by the use of the face recognition system. This system will
focus on face recognition in criminal investigations
It can allows the enhancement of images for comparison to the gallery.
This allows operators to develop watch lists of potential matches while
maintaining a full audit trail for each step in the process. It also helps
investigators identify individuals in crime scene photos and surveillance
videos by matching facial images against the agencys mugshot repository.
Examiner also provides a set of inspection tools that helps identify the
person in question in a timely manner, allowing investigators to act upon
the search results in the critical time period after a crime has been
committed.
Examiner enhances crime scene video face images for interlaced video,
poor lighting, motion blurriness, low resolution, aspect ratio inconsistencies
and off angle poses. The enhanced images are then searched against the
full Mugshot Repository with possible demographic filtering and candidate
lists created.
The system also contains the face search image verification tools (side
by-side split screens with interactive image viewing controls, key face
landmark feature angle and distance measurement tool) necessary for the
examiner to compare the search and candidate list face images to verify
investigation results.

2.2 Limitations and Assumptions


Research has shown that it is surprisingly difficult to match a face to an
image. This routine task, performed hundreds of times every day by
passport officers,

Security guards and police officers, turns out to be highly error prone.
In one of the experiment of this phenomenon a group of researcher did a
field test to establish the level of fraud protection afforded by the inclusion
of ID photos on credit cards. Supermarket checkout staff were required to
validate the photo-credit cards by deciding whether or not the photograph
was of the person presenting the card. Even though the staff were aware
they were taking part in a study concerning the utility of photo credit cards,
they performed surprisingly badly, with about half of the fraudulent cards
being accepted, and about one in 10 of the valid cards being falsely
rejected.
Here are some of the limitation or problems found in the system :
a) The resolution of the video images are not sufficient for facial
recognition algorithms
b) The data elements are not marked correctly on the sample data
c) The lighting and subject position in the video lend itself to photographic
issues such as perspective distortions, angle of pose issues, etc.
d) Statistical error analysis are missing in the expert witness work.
e) The defendant expert is not an expert in photography and Photoanthropometric area.
f) The data points used are not reliable points.
Hence, we could assume that that every criminals will have to go through a
series of high quality mugshot photo shoot to compare with its own
database. Second assumption is that the images captured will be converted
into grayscale images. This will help the system to overcome the risk of the
criminals changing their appearance.

3. Characteristics
Facial recognition system is based on the ability to recognize a face and then
measure the various features of the face. Every face has numerous,
distinguishable landmarks, the different peaks and valleys that make up

facial features. These landmarks defined as nodal points. Each human face
has approximately 80 nodal points. The features measured by the system are:
Distance between the eyes
Width of the nose
Depth of the eye sockets
The shape of the cheekbones
The length of the jaw line
These nodal points are measured creating a numerical code, called
a faceprint, representing the face in the database. But this system will focus
on eyes.

The eye, as in the iris, is the part of the face that is hard to change. For
example, part of face such as the nose, ears, shape of the face or even the
skin color could be changed by undergoing cosmetic surgery. By undergoing
this procedure, these criminals would be able to fool the system if the
proposed system focuses on these features, except for the iris. The criminal
could wear contact lenses to change their colour of their eye, but as we are a
step ahead by doing the recognition in grayscale, the system will be able to
recognise the criminal regardless of the colour of the eye.
.The distance between the eyes (iris) is able to increase the accuracy of the
system. When the system is retrieving the image, it will automatically measure
the distance between the irises, and will try to match with the database. If
match is found, added with the match of right or left eye, the system will notify
the law enforcement.

4. Suitable ANN for face recognition system


Artificial neural network is to be designed and trained to recognize the
persons of the database. An imaging system that converts each face image

centered in the systems field of vision is available.However the imaging


system may not be perfect and the faces may suffer from noise.
Perfect classification of ideal input images is required, and reasonably
accurate classification of noisy images. Hence some basic neural network
models
must be applied to each models in modules of face recognition system.

For face detection module, a three-layer feed forward artificial neural network
can be used detect human faces so that face detecting rate is rather high.
For face alignment module, a multilayer perceptron (MLP) with linear function
(three-layer) is used, and it creates 2D local texture model for the active
shape model (ASM) local searching. For feature extraction module, a method
for combination of geometric feature-based method and ICA method in facial
feature extraction is used. For
face matching, a model which combines many artificial neural networks
applied for geometric features classification is used.

5.1 Face recognition application


5.1.1 Input

10

To create a neural and train the system, images from the will be used. To
do this the network will first be trained on ideal images until it has a low
sum-squared error. Then the network will be trained on 8 sets of ideal
images. All training is done using back propagation with both adaptive
learning rate and momentum. All training sets are in grayscale to increase
accuracy. Here are some examples of the images of a person:

All the images are not the similar even though these picture are of the same
person. By doing this, the system will be able to adapt if there might be a
slight changes if the system is trying to recognize the same person but maybe
from a different angle. The position of the iris of the person is also varies as
this will help to train the system. The training example will also include person
with different features such as wearing spectacles, have facial hair, moles and
many more. This will help to train the system in better ways for detail match
inspection.
This will provides toolset for comparing the match results conveniently in
ranked order to confirm the specific match. Probe and reference images can
be examined in different views allowing for a more comprehensive match
confirmation.

For example:
The image to the right displays the side by side image verification screen with
the angle and distance measurement tool from Examiner used to track the
location of a facial mole location on the probe and file candidate images.

11

Below here is how the system read the image through Matlab.

% read images for i = from:length

data1{i-4} = imread(['C:\images\f1\' num2str(i) '.tif']); data1{i-4} =


imnoise(data1{i-4},'salt & pepper',u);
data1{i-4} = imadjust(data1{i-4},[lowin highin],[lowout highout]); end;

Below is the systems set epoch value

%-------------------\\\Train test set///----------------

hd=10;
lr=.25;
ep=20;

load wintface;load zintface; WX=winit;ZX=zinit;

save wq WX ; save zq ZX;

12

%--------------------------

To avoid increased computation and long training times, it is necessary to


resize the face images.

Therefore the image is resized30topixels30 immediately after

they have passed through the hit and miss kernels - that is, before entering
the classification stage. Since each subject has an equal number of 4 face
images in the training set and the test set, their input dimension to the system
is the900same:columns4rowspixel units. Each subject is trained
individually

Code that will resize the pictures.

%-------------------\\\Resize///----------------

rez = 30;
for i = from:length

temp = imread(['C:\images\f1\' num2str(i) '.tif']); temp = imresize(temp, [rez


rez]);
data1(i,:) = temp(:); end;
%--------------------------

First step is to run and train all the training set.

13

Then, second step is we train the test set

Interface

14

5.1.2 Process
After the system has been trained we can start process the images to be
detected.

for example if we choose no 6:

15

here is the detected image with recognition accuracy = 90%

16

Process on to train the eyes on the training set (Training Set)

(Validation Set)

train eye using (test set)

17

5.1.3 Output

The pre-recognition and actual recognition procedures are the same as for face images,
except that when an eye is detected, the system will retrieve the face to which that
detected eye belongs, together with the face from which the test eye image is cropped.

Assume that the system now have a detected face. The next step is to use the cropped
eye images to detect the eyes of the detected face. On this step, the detected face
image is loaded and convolution is performed using the cropped pair of left and right
eyes as templates. The detected eye region is indicated by a TAP (target aim point)
rectangle that surrounds it.

Next, the system will locate the eye level using Hough Transform. Then, the system will
shift the TAP rectangles to the correct position as final adjustments are made. The
output images are given as follow.

The system then will calculate the euclians distance between the eyes.

18

This figure shows the basic process on how the process of image recognition has been
done using artificial neural network.

19

6. Conclusion

Face recognition is challenging problems and there is still a lot of work that needs to be
done in this area. Over the years, face recognition has received substantial attention
from researchers in biometrics, pattern recognition, computer vision, and cognitive
psychology communities.This common interest in facial recognition technology among
researchers working in diverse fields is motivated both by the remarkable ability to
recognize people and by the increased attention being devoted to security applications.
Applications of face recognition can be found in security, tracking, multimedia, and
entertainment domains.
We have demonstrated how a face recognition system can be designed by artificial
neural network. Note that the training process did not consist of a single call to a training
function. Instead, the network was trained several times on various input such as image
of the same person that varies like different angles, face features(bear,spectacles and
many more).

Further work can be expanded in many ways.The algorithm can be extended to include
other recognition procedures, such as detecting faces from a group photograph or
faces in motion. In addition of this work, we can use Three-dimensional face
recognition (3D face recognition) instead of 2D images which is a modality of facial
recognition methods in which the three-dimensional geometry of the human face is used
which can increase the system accuracy.

20

7. References
1. http://www.nec.com/en/global/solutions/security/technologies/face_recognition
.html
http://ubiquity.acm.org/article.cfm?id=985625
http://www.priv.gc.ca/information/research-recherche/2013/fr_201303_e.asp
http://www.fbi.gov/about-us/cjis/fingerprints_biometrics/biometric-center-ofexcellence/modalities/facial-recognition
http://en.wikipedia.org/wiki/Facial_recognition_system