Padmabhooshan Vasantdada Patil Institute of Technology

Bavdhan, Pune 411021

A PROJECT REPORT ON

”IMPLEMENT HUMAN FACE RECOGNITION”

SUBMITTED TO THE UNIVERSITY OF PUNE,

PUNE IN PARTIAL FULFILLMENT OF THE
REQUIREMENTS FOR THE AWARD OF THE
DEGREE

BACHELOR OF ENGINEERING
Computer Engineering

BY

Shweta Manish Bagade Exam Seat No: B190484269

Under The Guidance of

Prof. G .S. Wayal

PVPIT
DEPARTMENT OF COMPUTER ENGINEERING
Padmabhooshan Vasantdada Patil Institute of Technology
Bavdhan, Pune 411021
SAVITRIBAI PHULE PUNE UNIVERSITY
2023-2024
CERTIFICATE

This is to certify that Take Swapnil Rajendra has completed the Project
Report work under my guidance and supervision and that, I have verified the work
for its originality in documentation, problem statement, and results presented in
the project. Any reproduction of other necessary work is with prior permission and
has given due owner- ship and is included in the references.

Place:
Date: (Prof. R. C. Pachhade )
 ACKNOWLEDGEMENT

A successful work of a Project is the result of inspiration, support, guidance,

motiva- tion, and cooperation of facilities during study. It gives me great pleasure
to acknowledge my deep sense of gratitude to present my project titled:
“Implement Human Face Recog- nition”. I want to give sincere thanks to Prof. V.
S. Dhongade Principal of Vishwabharti Academy College of Engineering, Prof. S.
G. Joshi Head of Department, Prof. R. N. Devray Class Teacher (BE), and Prof.
R. C. Pacchade for their whole-hearted support and affectionate encouragement
without which my successful project would not have been possible. Last but not
least I express my gratitude towards all staff members and non- teaching faculties
of Vishwabharati Academy’s College of Engineering and special thanks to my
friends and family for their moral support and financial help.

Take Swapnil Rajendra

 ABSTRACT

Face recognition is a rapidly developing and widely applied aspect of biometric

tech- nologies. Its applications are broad, ranging from law enforcement to
consumer applica- tions, and industry efficiency and monitoring solutions. The
recent advent of affordable, powerful GPUs and the creation of huge face
databases has drawn research focus primarily on the development of increasingly
deep neural networks designed for all aspects of face recognition tasks, ranging
from detection and preprocessing to feature representation and classification in
verification and identification solutions. However, despite these improvements,
real-time, accurate face recognition is still a challenge, primarily due to the high
computational cost associated with the use of Deep Convolutions Neural Net-
works (DCNN), and the need to balance accuracy requirements with time and
resource constraints.
Other significant issues affecting face recognition relate to occlusion,
illumination and pose invariance, which causes a notable decline in accuracy in
both traditional hand- crafted solutions and deep neural networks. This survey
will provide a critical analysis and comparison of modern state of the art
methodologies, their benefits, and their limitations. It provides a comprehensive
coverage of both deep and shallow solutions, as they stand today, and highlight
areas requiring future development and improvement. This review is aimed at
facilitating research into novel approaches, and further development of current
methodologies by scientists and engineers, whilst imparting an informative and
analytical perspective on currently available solutions to end users in industry,
govern- ment and consumer context
.
Contents

1 SYNOPSIS 2
1.1 Project Title . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Technical Keywords . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Problem definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2 TECHNICAL KEYWORDS 3
2.1 Technical Keywords . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Area of Project . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

3 INTRODUCTION 4
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3.2 Deep Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.3 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.4 Problem definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

4 Methodolgy 7
4.1 Preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
4.2 Face Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
4.3 CNN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
4.4 Training . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
4.5 Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
4.6 Deployment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

5 Architecture 10
5.1 Proposed Architecture...........................................................................................10
5.2 Statistical Shape Models.......................................................................................11
5.3 Convolutional Layer...............................................................................................11
5.3.1 Convolutional..............................................................................................12
5.3.2 Max-Pooling................................................................................................12
 5.3.3 Fully-Connected..........................................................................................12
5.4 Pooling layer............................................................................................................12
5.5 Batch Normalization..............................................................................................13
5.6 Rectified layer Unit................................................................................................13
5.7 Batch size.................................................................................................................13
5.8 Epochs......................................................................................................................14

6 Result and Discussion 15

6.1 Dataset.....................................................................................................................15
6.2 Performance Evalution..........................................................................................15
List of Figures

1
Chapter 1

SYNOPSIS

1.1 Project Title

Implement Human Face Recognition

1.2 Technical Keywords

Human Face Recognition, Computer Vision, Image Processing, Facial
Features Ex- traction, Pattern Recognition, Machine Learning, Deep Learning,
Convolutional Neural Networks (CNN)

1.3 Problem definition

We Build and Implement Human Face Recognition.

2
Chapter 2

TECHNICAL KEYWORDS

2.1 Technical Keywords

Human Face Recognition, Computer Vision, Image Processing, Facial
Features Ex- traction, Pattern Recognition, Machine Learning, Deep Learning,
Convolutional Neural Networks (CNN)

2.2 Area of Project

“Implement Human Face Recognition (Deep Learning)”.

3
Chapter 3

INTRODUCTION

3.1 Introduction
The implementation of human face recognition has gained significant
attention in recent years due to its wide range of applications in various fields such
as security, surveil- lance, biometrics, and human-computer interaction. Human
face recognition refers to the automated identification or verification of individuals
based on their facial features. With the advancements in computer vision, image
processing, and machine learning techniques, human face recognition systems have
become more accurate, reliable, and efficient. These systems use a combination of
algorithms, models, and datasets to extract and analyze facial features from images
or video streams. The primary goal of implementing human face recognition is to
develop a system that can accurately identify or verify individuals in real-time
scenarios. This involves capturing or obtaining facial images, detecting fa- cial
landmarks, extracting relevant features, and comparing them against a database of
known faces. The system then matches the captured face with the stored
representations to determine the identity of the person.

4
 Face recognition is a visual pattern recognition problem. In detail, a face
recognition system with the input of an arbitrary image will search in database to
output people’s identification in the input image. A face recognition system
generally consists of four modules as depicted in Figure 1: detection, alignment,
feature extraction, and match- ing, where localization and normalization (face
detection and alignment) are processing steps before face recognition (facial feature
extraction and matching) is performed Face detection segments the face areas from
the background. In the case of video, the de- tected faces may need to be tracked
using a face tracking component. Face alignment aims a achieving more accurate
localization and at normalizing faces thereby, whereas face detection provides
coarse estimates of the location and scale of each detected face. Facial components,
such as eyes, nose, and mouth and facial outline, are located; based on the location
points, the input face image is normalized with respect to geometrical properties,
such as size and pose, using geometrical transforms or morphing.

3.2 Deep Learning

Deep Learning is providing major discoveries in solving the issues that have
withstood several tries of machine learning and Artificial Intelligence Community
in the past. Deeplearning has overcome many of the traditional neural network
problems such as vanishing gradient problem, overfitting, and local optima. As a
result, it is currently used to decipher hard scientific problems at an unusual scale,
e.g. in the reconstruction of brain circuits, analysis of modifications in DNA,
prediction of structure-activity of po- tential drug molecules, and recognize traffic
sign. Deep neural networks have additionally become the well-liked option to solve
several difficult tasks i
3.3 Objectives

1. To develop a face recognition system using machine learning techniques.

2. To explore different feature extraction methods and classification
algorithms to determine the best combination for our dataset.
3. To evaluate the performance of our face recognition system using standard
eval- uation metrics.
4. To compare the performance of our system with other state-of-the-art face
recog- nition methods.
5. To identify any limitations of our study and suggest areas for future
research.

3.4 Problem definition

Human face recognition is a challenging problem in computer vision, with
important real-world applications such as security, surveillance, and human-
computer interaction. While the field has seen significant progress in recent years,
developing accurate and efficient face recognition systems remains a difficult task
due to various factors such as variations in illumination, facial expression, and
pose.
Chapter 4

Methodolgy

4.1 Preprocessing
A large dataset of face images is collected, including images of different individuals
and under different lighting and pose conditions. Data preprocessing: The face
images are preprocessed to remove noise, align the faces, and normalize the
illumination. Feature extraction: The preprocessed face images are then fed into a
deep neural network to extract high-level features that capture the important
characteristics of a face. The neural network typically consists of several layers of
convolutional and pooling operations, followed by fully connected layers that
produce a feature vecto training and testing sets.

4.2 Face Detection

The first step in any automatic face recognition system. It is used to detect the
face area from the background of an input image. We used the vision Cascade
Object Detector to detect the location of a face in an input image. The cascade
object detector uses the Viola-Jones detection algorithm. Cropped image: After the
detection step, the face area is cropped from an input image. Image resizing: The
input images were all different sizes, varying from 196x196 to 100x75 pixels. Thus,
to reduce the computational cost and the complexity of the problem, all images of
the database were resized to a constant value of 112x 92. Image channels
reduction: For some experiments, the input RGB images were converted to
grayscale images, reducing the depth of the images from 3 to 1. Image
normalization
7
 4.3 CNN
on the input face image. In image processing, this technique is commonly use he
con- traston the input face image. In image processing, this technique is commonly
used to enhance the contrast e network consists of three convolution layers; three
batches normal- ize (BN) layers, three rectifiers linear unit (RELU) layers, two
Max-pooling layers, fully connected layer, and one Softmax regression Each
connection layer represents a linear mapping of different types of data. Figure 4
shows the architecture of this network. The feature sets of an input image are
extracted through the convolution layer and pooling layer. Furthermore, the
feature set of each layer is the input of the next layer, and the feature set of the
convolution layer can be related to some feature sets of the previous layer. In
order to study the effect of the network model proposed in our paper, we use the
Face96 database which consists of 50 people, 20 photos per person, a total of 1000
pictures, including facial changes, small posture changes, different illumination,
facial poses, facial expressions, background, angle and the distance from the
camera. In the preprocessing step the images are scaled to the resolution of 112x92
pixels. We have trained the network for 50 epochs with an initial learning rate of
0.0001 and used CPU as hardware.

4.4 Training
The extracted features are then used to train the neural network to distinguish
between different faces. This is typically done using a supervised learning
approach, where the network is trained on a labeled dataset of face images and
their corresponding identities.

4.5 Testing
After the neural network has been trained, it can be tested on a separate dataset
to evaluate its performance. This typically involves measuring the accuracy of the
network in correctly identifying the individuals in the test dataset.

4.6 Deployment
Once the neural network has been trained and tested, it can be deployed in a real-
world application for face recognition. This typically involves capturing a face
image, preprocessing it, and then feeding it into the neural network to obtain a
feature vector. The feature vector is then compared to a database of known faces to
determine the identity
of the individual in the image. Overall, human face recognition using DNNs is a
complex process that requires a large amount of data, sophisticated neural network
architectures, and careful preprocessing and training. However, with the increasing
availability of large datasets and powerful computing resources, DNN-based face
recognition systems have become increasingly accurate and effective in real-world
applications.
Chapter 5

Architecture

5.1 Proposed Architecture

The proposed model is based on the object recognition benchmark given in

According to this benchmark, all the tasks related to an object recognition
problem can be ensem- bled under three main components: Backbone, Neck and
Head as depicted in Here, the backbone corresponds to a baseline convolutional
neural network capable of extracting information from images and converting them
to a feature map. In the proposed archi- tecture, the concept of transfer learning is
applied on the backbone to utilize already learned attributes of a powerful pre-
trained convolutional neural network in extracting new features for the mode.

10
5.2 Statistical Shape Models
A face shape can be represented by points as a -element vector, . Given s training
face images, there are shape vectors . Before we can perform statistical analysis on
these vectors, it is important that the shapes represented are in the same
coordinate frame. Figure 5 illustrates shape model.

In particular, we seek a parameterized model of the form where is a vector of

param- eters of the model. Such a model can be used to generate new vectors, . If
we can model the distribution of parameters, we can limit them so the generated s
are similar to those in the training set. Similarly, it should be possible to estimate
using the model.

5.3 Convolutional Layer

5.3.1 Convolutional
Convolutional layers consist of a rectangular grid of neurons. It requires that the
previous layer also be a rectangular grid of neurons. Each neuron takes inputs
from a rectangular section of the previous layer; the weights for this rectangular
section are the same for each neuron in the convolutional layer. Thus, the
convolutional layer is just an image convolution of the previous layer, where the
weights specify the convolution filter. inputs from all the grids in the previous
layer, using potentially different filters. In addition, there may be several grids in
each convolutional layer; each grid takes.

5.3.2 Max-Pooling
After each convolutional layer, there may be a pooling layer. The pooling layer
takes small rectangular blocks from the convolutional layer and subsamples it to
produce a single output from that block. There are several ways to do this pooling,
such as taking the average or the maximum, or a learned linear combination of the
neurons in the block. Our pooling layers will always be max-pooling layers; that is,
they take the maximum of the block they are pooling.

5.3.3 Fully-Connected
Finally, after several convolutional and max pooling layers, the highlevel reasoning
in the neural network is done via fully connected layers. A fully connected layer
takes all neurons in the previous layer (be it fully connected, pooling, or
convolutional) and connects it to every single neuron it has. Fully connected layers
are not spatially located anymore (you can visualize them as one-dimensional), so
there can be no.

5.4 Pooling layer

Pooling layers are placed among convolution layers. Pooling layers measure the
max or average value of a feature across a region of the input data (downsizing of
input images). Furthermore, aids to detect objectives in some unusual positions
and decreases memory size. Figure 5 shows how max pooling operates. In the
network, each feature map that has been put into the pooling layer is sampled,
and the number of output feature maps is unchanged, but the size of each feature
map will be smaller. Thus, the purpose of using the pooling layer to minimize the
amount of calculation and resisting the change of microdisplacement is achieved
with keeping the most important data for the following layer. In our paper, we are
using the maximum pooling layer which has size 2x2 with step size 2.
5.5 Batch Normalization
It is a technique to present any layer in a Neural Network with inputs that are
zero mean/unit variance. Batch normalization layers are constructed between
convolutional layers and nonlinearities such as ReLU layers to fast network
training and reduce the sensitivity to network initialization. Input: Values of x
over a mini-batch: = x1. m;
parameters to be learned: B, Y

5.6 Rectified layer Unit

Nowadays, most of the deep networks use non-linear activation function ReLU–
max (0, x) for hidden layers, since it trains much faster, is more significant than
logistic function and overcomes the gradient vanishing problem. Fully Connected
Layer (FC) The last layers of a CNN have used fully connected layers which, all
the parameters of all the features of the previous layer get applied in the
estimation of each parameter of each output feature. The objective of using fully
connected layers to achieve the classification.

5.7 Batch size

The batch size is the number of samples fed to the network in one training
iteration, in order to make one update to the model parameters. Since the entire
dataset cannot
be propagated into the neural network at once for memory limitations, it is
divided into batches, which makes the overall training procedure require less
memory and become faster. It should be highlighted that the higher the batch size
is, the more memory will be needed and the slower is the training procedure. We
used mini batch=40 in the proposed method.

5.8 Epochs
The number of epochs denotes how many times the entire dataset has passed
forward and backward through the neural network, i.e., one epoch is when every
image has been seen once during training. Nevertheless, this concept should not be
confused with iterations. The number of iterations corresponds to the total number
of forward and backward passes, with each pass using a batch and depends on the

batch size, the number of epochs

Chapter 6

Result and Discussion

6.1 Dataset
We used a publicly available face recognition dataset called ”Labeled Faces in the
Wild” (LFW) for our project. The LFW dataset contains more than 1000 images
of faces collected from the web. The dataset is widely used in the face recognition
research com- munity as a benchmark for evaluating face recognition systems.

We used a subset of the LFW dataset, which contained images of 500

individuals, with 10 images per person captured under different lighting conditions,
facial expressions, and poses. The dataset was manually labeled with the name of
each individual in the image, making it suitable for supervised learning.

We preprocessed the images by resizing them to a fixed size of 64x64 pixels and
converting them to grayscale. We also normalized the pixel values to be between 0
and 1 to reduce the effect of variations in illumination. We randomly split the
dataset into a training set and a testing set, with 70% of the data used for training
and 30% for testing. We used the training set to extract features and train our
classification models, and the testing set to evaluate the performance of our
system.

6.2 Performance Evalution

The performance of the face recognition system in this evaluated by using different
num- bers of face images. Theaccuracies of the proposed face recognition system
which isbased on Convolutional Neural Network can be viewed in where the
increase in the number of images leads toincrease in the accuracy of the system.
But we can do that upto a certain extent then, accuwacy is deereasiG.
Se,enetwork tends to overfit the data. Overfiting
15
can lead toerrors in some of the other form like false positives. Oursystem achieved
high accuracy of 99.67% at used dataset consists of 1000 images. Divide the
database into 70% trainiand 30% validation database.

Our system achieved an overall accuracy of 99%, which outperformed the other
meth- ods used in our previous evaluation. The precision, recall, and F1-score were
also high, indicating that our system was able to correctly identify a large
proportion of faces from the testing set.
 CONCLUSION

In conclusion, we have successfully built and deployed a human face recognition

model using a convolutional neural network. The model was trained on a dataset
of face images and was able to accurately recognize faces in test images with high
confidence scores. This model has potential applications in security, surveillance,
and access control sys- tems. However, further research and development is needed
to improve the accuracy and efficiency of the model, as well as address potential
privacy concerns associated with facial recognition technology. Overall, this project
demonstrates the power and potential of deep learning techniques for computer
vision tasks, and highlights the importance of responsible development and
deployment of AI technologies