An

Industrial Oriented Mini Project Report

on
TRUTHLENS: DEEPFAKE DETECTION THROUGH THE LENS OF DEEP
LEARNING
submitted in partial fulfillment of the requirement for the award of degree

BACHELOR OF TECHNOLOGY

In

COMPUTER SCIENCE AND ENGINEERING

Of

JAWAHARLAL NEHRU TECHNOLOGICAL UNIVERSITY- HYDERABAD

Submitted By
MARELLA SAMARASIMHA REDDY 215U1A05B4
MAKALA RAMANJANEYULU 215U1A05B2
KOLLURI RAJENDHER 215U1A0595
MUDAVATH BHARGAVI 215U1A05C5

Under the guidance of

Dr. B. Shilpa Reddy

Assistant Professor

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

AVN INSTITUTE OF ENGINEERING AND TECHNOLOGY

Approved by AICTC, Accredited by NAAC & NBA, Affiliated to JNTUH
Koheda Road, M.P.Patelguda Post, Ibrahimpatnam (M),Ranga Reddy Dist, T.G.S., India.

2024-25

i
 CERTIFICATE

This is to certify that the project report entitled TruthLens: Deepfake Detection
Through the Lens of Deep Learning submitted by Marella Samarasimha
Reddy, Makala Ramanjaneyulu, Kolluri Rajendher and Mudavath Bhargavi
to the AVN Institute of Engineering & Technology, in partial fulfillment for the
award of the degree of B. Tech in Computer Science and Engineering is a bona
fide record of mini project work carried out by them under our supervision. The
contents of this report, in full or in parts, have not been submitted to any other
Institution or University for the award of any degree or diploma.

Guide Project Coordinator HOD

Dr. B. Shilpa Mr. M. Kanakeshwar Reddy Dr. V. Goutham
Assistant Professor Assistant Professor Professor

External Examiner

ii
 DECLARATION

We declare that this project report titled “TruthLens: Deepfake

Detection Through the Lens of Deep Learning” submitted in partial
fulfillment of the degree of B. Tech in Computer Science and
Engineering is a record of original work carried out by us under the
supervision of Dr. B. Shilpa Reddy and has not formed the basis for
the award of any other degree or diploma, in this or any other
Institution or University. In keeping with the ethical practice in
reporting scientific information, due acknowledgements have been
made wherever the findings of others have been cited.

Signature

Marella Samarasimha Reddy 215U1A05B4

Makala Ramanjaneyulu 215U1A05B2

Kolluri Rajendher 215U1A0595

Mudavath Bhargavi 215U1A05C5

iii
 ACKNOWLEDGMENTS
It gives us great pleasure to present the report of the Mini Project Work
undertaken during B.Tech. Final Year. We owe a special debt of gratitude to our
Project Guide Dr. B. Shilpa Reddy, Assistant Professor, Department of Computer
Science and Engineering, for their constant support and guidance throughout the
course of our work. It is only through their efforts that our endeavors have seen
the light of day.
Our deepest thanks to our Project Coordinator Ms. Pooja Reddy,
Assistant Professor, Department of Computer Science and Engineering, for
guiding and correcting various documents of us with attention and care.
We also take this opportunity to acknowledge the contribution of Dr.
V Goutham, Head of the Department of Computer Science and Engineering for
his full support and assistance during the development of the project.
We would also like to acknowledge the contribution of all faculty
members of the department for their kind assistance and cooperation during the
development of our project.

Marella Samarasimha Reddy 215U1A05B4

iv
Makala Ramanjaneyulu 215U1A05B2
Kolluri Rajendher 215U1A0595
Mudavath Bhargavi 215U1A05C5

v
 ABSTRACT

With the rise of AI-generated media, particularly deepfakes—synthetic images

where a person’s likeness is digitally manipulated—ensuring image authenticity has
become a critical challenge in digital media forensics. This project focuses on
detecting deepfake images using a Convolutional Neural Network (CNN) to classify
them as either real or fake based on subtle visual artifacts present in the images. The
detection leverages key facial features. First, facial geometry and alignment are
analyzed to identify disproportionate face shapes or misaligned facial features, which
are common in deepfakes. Second, inconsistencies in eye reflections serve as
indicators of fake content. The model also detects irregularities in mouth and teeth
regions, such as blurry teeth or unnatural renderings. Lastly, mismatches in shadows
and lighting, particularly around the face, further help identify deepfake images.
Using an extensive dataset of authentic and deepfake images, this project
preprocesses the data by normalizing the images and extracting facial regions. The
CNN model is trained to focus on these informative regions, enabling it to classify
images accurately. By learning to identify manipulated regions, the model provides a
reliable method for deepfake detection and offers a significant contribution to the
growing concerns surrounding synthetic media on social platforms.

v
LIST OF FIGURES

FIGURE TITLE PAGE

NUMBER
3.1 Proposed model of Deepfake detection 8

3.2 Data pre-processing 10

3.3 ResNet50 architecture 12

3.4 EfficientNet-B0 14

3.5 MobileNetv2 16

4.1 Sequence diagram 23

4.2 Activity Diagram 26
4.3 Streamlit interface 29
4.4 Choosing an image 29
4.5 Uploaded image 30
4.6 Prediction 31
5.1 Training and testing results 33

LIST OF ABBREVIATIONS
vi
CNN : Convolutional Neural Network
ResNet50 : Residual Network 50 layers
EfficientNet-B0 : Efficient Network B0 (a baseline variant of EfficientNet)
MobileNetV2 : Mobile Network Version 2
Dlib : C++ Toolkit for Machine Learning and Computer Vision
PyTorch : Python-based deep learning framework
GPU : Graphics Processing Unit
CUDA : Compute Unified Device Architecture (NVIDIA's parallel
computing platform and programming model)
IoU : Intersection over Union (used for evaluating object detection
performance)
ROC AUC : Receiver Operating Characteristic - Area Under Curve (a
performance metric)
F1 Score : The harmonic mean of precision and recall, used for classification
model evaluation
Precision : The ratio of correctly predicted positive observations to the total
predicted positives
Recall : The ratio of correctly predicted positive observations to the total
actual positives
TP : True Positive
TN : True Negative
FP : False Positive
FN : False Negative
VAE : Variational Autoencoder (used in generative models)
GAN : Generative Adversarial Network
WGAN : Wasserstein Generative Adversarial Network
cGAN : Conditional Generative Adversarial Network
FC : Fully Connected (layer)
CVPR : Computer Vision and Pattern Recognition (Conference)

vii
IEEE : Institute of Electrical and Electronics Engineers

viii
 TABLE OF CONTENTS

DESCRIPTION PAGE NUMBER

CERTIFICATE ii
DECLARATION iii
ACKNOWLEDGEMENTS iv
ABSTRACT v
LIST OF FIGURES vi
ABBREVIATIONS vii
1. INTRODUCTION
1.1 Motivation 1
1.2 Problem Statement 1
1.3 Purpose 2
1.4 Scope 2
1.5 Project Objective 3
2. LITERATURE REVIEW
2.1 Overview of Related Work 4
2.2 Studies and Research 4
2.3 Identifying Gaps and Opportunities 6
3. METHODOLOGY
3.1 Proposed Method 8
3.2 Data Collection 9
3.3 Data Pre-Processing 9
3.4 System Design 10
3.4.1 Resnet50 11
3.4.2 EfficientNet-B0 13
3.4.3 MobileNetV2 15
3.5 Algorithms and Techniques 17
3.5.1 Face Detection and Cropping 17
3.5.2 Data Augmentation 17
3.5.3 Feature Extraction and Transfer Learning 18
 3.5.4 Ensemble Modeling 19
3.5.5 Loss Function 19
3.5.6 Optimization and Learning Rate Scheduling 20
3.5.7 Early Stopping 20
4. IMPLEMENTATION
4.1 Development Environment and Tools 21
4.2 Sequence Diagram 23
4.3 Activity Diagram 26
4.4 Streamlit Implementation 28
4.5 User Interface Design 28
5. RESULTS AND ANALYSIS
5.1 Results and Discussion 32
6. CONCLUSION AND FUTURE SCOPE
6.1 Conclusion 34
6.2 Future Scope 35
6.2.1 Dataset Expansion and Diversity 35
6.2.2 Extending to Video Based Detection 35
REFERENCES 36
APPENDIX I Code 37