A PRELIMINARY PROJECT REPORT ON

DEEPFAKE VIDEO DETECTION

SUBMITTED TO THE SAVITRIBAI PHULE PUNE UNIVERSITY , PUNE IN THE

PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE AWARD OF THE
DEGREE

BACHELOR OF ENGINEERING
(Computer Engineering)
BY
Nikhil Singh Joadha Exam No:
Khushi Solanki Exam No:
Tejas Mehtre Exam No:
Gitesh Deshmukh Exam No:

Under The Guidance of

Dr. K. C. NALAVADE

DEPARTMENT OF COMPUTER ENGINEERING SANDIP

INSTITUTE OF ENGINEERING & MANAGEMENT
MAHIRAVANI, TRIMBAK ROAD, NASHIK 422213

SAVITRIBAI PHULE PUNE UNIVERSITY, PUNE

2024 - 2025
 CERTIFICATE

This is to certify that the Project Entitled

Deep Fake Video Detection
Submitted by
Nikhil Singh Jodha Exam No:
Khushi Solanki Exam No:
Tejas Mehtre Exam No:
Gitesh Deshmukh Exam No:
is a bonafide work carried out by them under the supervision of Dr.K.C. NALAVADE
and it is approved for the partial fulfillment of the requirement of Savtribai Phule Pune
university, Pune for the award of the degree of Bachelor of Engineering (Computer En-
gineering).

Dr. K. C. Nalavade Dr. K. C. Nalavade

Guide H.O.D
Dept. of Computer Engg. Dept. of Computer Engg.

Dr. D. P. Patil
Principal
Sandip Institute of Engineering & Management

Signature of Internal Examiner Signature of External Examiner

 PROJECT APPROVAL SHEET

A Project Report Titled as

Deep Fake Video Detection

Is verified for its originality in documentation, problem statement, proposed work and
implementation successfully completed by

Nikhil Singh Jodha Exam No:

Khushi Solanki Exam No:
Tejas Mehtre Exam No:
Gitesh Deshmukh Exam No:

at

DEPARTMENT OF COMPUTER ENGINEERING

SANDIP INSTITUTE OF ENGINEERING & MANAGEMENT

SAVITRIBAI PHULE PUNE UNIVERSITY,PUNE

ACADEMIC YEAR 2019-2020

Dr. K. C. Nalavade Dr. K. C. Nalavade

Guide H.O.D
Dept. of Computer Engg. Dept. of Computer Engg.
 Abstract

The rapid advancement of artificial intelligence has fueled the creation of deepfake
videos, posing serious ethical and security challenges. These videos use AI algorithms to
manipulate visual and auditory content, creating hyper-realistic media that can mislead
and deceive audiences. This project aims to develop a robust deepfake detection system
using a hybrid approach that leverages Convolutional Neural Networks (CNN), Recur-
rent Neural Networks (RNN) with Long Short-Term Memory (LSTM), and Generative
Adversarial Network (GAN) verification. By combining these models, our approach ef-
fectively captures both spatial and temporal features crucial for distinguishing between
authentic and manipulated videos.

The proposed solution begins with comprehensive preprocessing of video datasets to

enhance feature clarity and consistency, followed by a data split into training and testing
sets in a 70-30 ratio. To extract spatial features from individual frames, we employ
FFmpeg and OpenCV, enabling the model to focus on fine-grained details within each
frame. CNN captures spatial characteristics, while the RNN/LSTM component identifies
sequential dependencies, enhancing the model’s ability to detect subtle inconsistencies
over time. GAN verification further strengthens the model’s accuracy by simulating
potential manipulations, allowing the system to generalize to various deepfake styles.

The effectiveness of our model is evaluated using a confusion matrix, which assesses
metrics such as accuracy, precision, and recall. Results from experiments show that this
hybrid model significantly improves deepfake detection accuracy, outperforming tradi-
tional single-model approaches. This project’s hybrid framework has broad applications
in digital forensics, media authenticity verification, and social media monitoring, con-
tributing a powerful tool to combat the rising threat of deepfake media.

Keywords: Deepfake Detection,Hybrid Model (CNN-LSTM-GAN),Spatial and Tem-

poral Feature Extraction,Confusion Matrix Evaluation

SIEM, Department of Computer Engineering 2019-20 I

 Acknowledgments

It gives us great pleasure in presenting the preliminary project report on ‘DeepFake

Video Detection’.

I would like to take this opportunity to thank my internal guide Dr. K. C. Nalavade
for giving me all the help and guidance I needed. I am really grateful to them for their
kind support. Their valuable suggestions were very helpful.

I am also grateful to Dr. K. C. Nalavade, Head of Computer Engineering Department,

Sandip Institute of Engineering & Management for his indispensable support, suggestions.

In the end our special thanks to Dr. D. P. Patil for providing various resources such as
laboratory with all needed software platforms, continuous Internet connection, for Our
Project.

Nikhil Singh Jodha

Khushi Solanki
Tejas Mehtre
Gitesh Deshmukh
(B.E. Computer Engg.)

SIEM, Department of Computer Engineering 2019-20 II

Contents

1 Synopsis 1
1.1 Project Title . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Project Option . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Internal Guide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.4 Sponsorship and External Guide . . . . . . . . . . . . . . . . . . . . . . 2
1.5 Technical Keywords (As per ACM Keywords) . . . . . . . . . . . . . . . 2
1.6 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.7 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.8 Goals and Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.9 Relevant Mathematics associated with the Project . . . . . . . . . . . . . 4
1.10 Names of Conferences / Journals where papers can be published . . . . . 4
1.11 Review of Conference/Journal Papers supporting Project idea . . . . . . 4
1.12 Plan of Project Execution . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2 Technical Keywords 6
2.1 Area of Project . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Technical Keywords . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

3 Introduction 8
3.1 Project Idea . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.2 Motivation of the Project . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.3 Literature Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

4 Problem Definition and scope 11

4.1 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.1.1 Goals and objectives . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.1.2 Statement of scope . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.2 Major Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.3 Methodologies of Problem solving and efficiency issues . . . . . . . . . . 13
4.4 Outcome . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4.5 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4.6 Hardware Resources Required . . . . . . . . . . . . . . . . . . . . . . . . 15
4.7 Software Resources Required . . . . . . . . . . . . . . . . . . . . . . . . . 15

5 Project Plan 16
5.0.1 Reconciled Estimates . . . . . . . . . . . . . . . . . . . . . . . . . 17
5.0.2 Project Resources . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.1 Risk Management w.r.t. NP Hard analysis . . . . . . . . . . . . . . . . . 18
5.1.1 Risk Identification . . . . . . . . . . . . . . . . . . . . . . . . . . 18
 5.1.2 Risk Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
5.1.3 Overview of Risk Mitigation, Monitoring, Management . . . . . . 19
5.2 Project Schedule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5.2.1 Project task set . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5.2.2 Task network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5.2.3 Timeline Chart . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5.3 Team Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5.3.1 Team structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5.3.2 Management reporting and communication . . . . . . . . . . . . . 22

6 Software requirement specification 23

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
6.1.1 Purpose and Scope of Document . . . . . . . . . . . . . . . . . . 24
6.1.2 Overview of responsibilities of Developer . . . . . . . . . . . . . . 24
6.2 Usage Scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
6.2.1 User profiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
6.2.2 Use-cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
6.2.3 Use Case View . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
6.3 Data Model and Description . . . . . . . . . . . . . . . . . . . . . . . . . 24
6.3.1 Data Description . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
6.3.2 Data objects and Relationships . . . . . . . . . . . . . . . . . . . 24
6.4 Functional Model and Description . . . . . . . . . . . . . . . . . . . . . . 26
6.4.1 Data Flow Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . 26
6.4.2 Activity Diagram: . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
6.4.3 Non Functional Requirements: . . . . . . . . . . . . . . . . . . . . 26
6.4.4 State Diagram: . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
6.4.5 Design Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . 26
6.4.6 Software Interface Description . . . . . . . . . . . . . . . . . . . . 26

7 Detailed Design Document using Annexure A and B 28

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
7.2 Architectural Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
7.3 Data design (using Annexures A and B) . . . . . . . . . . . . . . . . . . 29
7.3.1 Internal software data structure . . . . . . . . . . . . . . . . . . . 30
7.3.2 Global data structure . . . . . . . . . . . . . . . . . . . . . . . . . 30
7.3.3 Temporary data structure . . . . . . . . . . . . . . . . . . . . . . 30
7.3.4 Database description . . . . . . . . . . . . . . . . . . . . . . . . . 30
7.4 Compoent Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
7.4.1 Class Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

8 Project Implementation 32
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
8.2 Tools and Technologies Used . . . . . . . . . . . . . . . . . . . . . . . . . 33
8.3 Methodologies/Algorithm Details . . . . . . . . . . . . . . . . . . . . . . 33
8.3.1 Algorithm 1/Pseudo Code . . . . . . . . . . . . . . . . . . . . . . 33
8.3.2 Algorithm 2/Pseudo Code . . . . . . . . . . . . . . . . . . . . . . 33
8.4 Verification and Validation for Acceptance . . . . . . . . . . . . . . . . . 33

9 Conclusion and Future Scope 34

SIEM, Department of Computer Engineering 2019-20 IV

References 35

10 References 37

Annexure A Laboratory Assignments on Project Analysis of Algorithmic

Design 39

Annexure B Laboratory Assignments on Project Quality and Reliability

Testing of Project Design 41

Annexure C Project Planner 43

Annexure D Reviewers Comments of Paper Submitted 45

Annexure E Plagiarism Report 47

Annexure F Information of Project Group Members 49

List of Figures

4.1 System Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

5.1 Timeline Chart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

6.1 Use case diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

6.2 State transition diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

7.1 Architecture diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

7.2 Class Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

SIEM, Department of Computer Engineering 2019-20 VI

List of Tables

4.1 Hardware Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

5.1 Risk Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

5.2 Risk Probability Definitions . . . . . . . . . . . . . . . . . . . . . . . . . 20
5.3 Risk Impact Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
5.4 Risk for Development Environment . . . . . . . . . . . . . . . . . . . . . 21
5.5 Risk for Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
5.6 Risk for Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

6.1 Use Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

A.1 IDEA Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

CHAPTER 1
SYNOPSIS
1.1 Project Title
DeepFake Video Detection

1.2 Project Option

Internal Project

1.3 Internal Guide

Dr. K. C. Nalavade

1.4 Sponsorship and External Guide

NA

1.5 Technical Keywords (As per ACM Keywords)

1. Video Analysis

2. Machine Learning

3. Neural Networks

4. Computer Vision

5. Image Processing

6. Generative Adversarial Networks (GANs)

7. Temporal Dynamics

8. Artificial Intelligence

SIEM, Department of Computer Engineering 2019-20 2

1.6 Problem Statement
With the rise of deepfake technology, there is an crucial need for effective detection
methods to prevent the misuse of manipulated media in security, privacy, and information
integrity contexts. Current detection models struggle to generalize across varied deep-
fake types, and there is a need for approaches that leverage both spatial and temporal
features for improved accuracy. This project aims to develop a robust deepfake detection
system using a hybrid deep learning approach involving Convolutional Neural Networks
(CNN), Long Short-Term Memory (LSTM) networks, and Generative Adversarial Net-
work (GAN) verification. By leveraging FFmpeg for frame extraction and OpenCV for
spatial feature analysis, the model will be trained and tested on preprocessed datasets
split in a 70-30 format. The system’s effectiveness will be evaluated using a confusion
matrix to ensure reliable and accurate detection of deepfake content.

1.7 Abstract
The rapid advancement of artificial intelligence has fueled the creation of deepfake
videos, posing serious ethical and security challenges. These videos use AI algorithms to
manipulate visual and auditory content, creating hyper-realistic media that can mislead
and deceive audiences. This project aims to develop a robust deepfake detection system
using a hybrid approach that leverages Convolutional Neural Networks (CNN), Recur-
rent Neural Networks (RNN) with Long Short-Term Memory (LSTM), and Generative
Adversarial Network (GAN) verification. By combining these models, our approach ef-
fectively captures both spatial and temporal features crucial for distinguishing between
authentic and manipulated videos.

The proposed solution begins with comprehensive preprocessing of video datasets to

enhance feature clarity and consistency, followed by a data split into training and testing
sets in a 70-30 ratio. To extract spatial features from individual frames, we employ
FFmpeg and OpenCV, enabling the model to focus on fine-grained details within each
frame. CNN captures spatial characteristics, while the RNN/LSTM component identifies
sequential dependencies, enhancing the model’s ability to detect subtle inconsistencies
over time. GAN verification further strengthens the model’s accuracy by simulating
potential manipulations, allowing the system to generalize to various deepfake styles.

The effectiveness of our model is evaluated using a confusion matrix, which assesses
metrics such as accuracy, precision, and recall. Results from experiments show that this
hybrid model significantly improves deepfake detection accuracy, outperforming tradi-
tional single-model approaches. This project’s hybrid framework has broad applications
in digital forensics, media authenticity verification, and social media monitoring, con-
tributing a powerful tool to combat the rising threat of deepfake media.

Keywords: Deepfake Detection,Hybrid Model (CNN-LSTM-GAN),Spatial and Tem-

poral Feature Extraction,Confusion Matrix Evaluation

SIEM, Department of Computer Engineering 2019-20 3

1.8 Goals and Objectives
• To develop an efficient and accurate deepfake detection model.

• To evaluate the system on challenging datasets and compare its performance with
existing methods.

• To analyze and compare the performance of different deep learning architectures

for deepfake detection.

• To implement a user-friendly interface for real-time detection of deepfake videos.

1.9 Relevant Mathematics associated with the Project

System Description:

• Input:

• Output:

• Identify data structures, classes, divide and conquer strategies to exploit distributed/parallel/concu
processing, constraints.

• Functions : Identify Objects, Morphisms, Overloading in functions, Functional re-

lations

• Mathematical formulation if possible

• Success Conditions:

• Failure Conditions:

1.10 Names of Conferences / Journals where papers

can be published
• IEEE/ACM Conference/Journal 1

• Conferences/workshops in IITs

• Central Universities or SPPU Conferences

• IEEE/ACM Conference/Journal 2

1.11 Review of Conference/Journal Papers support-

ing Project idea
Atleast 10 papers + White papers or web references
Brief literature survey [ Description containing important description of at least 10 papers
1.12 Plan of Project Execution
Using planner or alike project management tool.

SIEM, Department of Computer Engineering 2019-20 5

 CHAPTER 2
TECHNICAL KEYWORDS
2.1 Area of Project
Artificial Intelligence and Machine Learning

2.2 Technical Keywords

1. Video Analysis

2. Machine Learning

3. Neural Networks

4. Computer Vision

5. Image Processing

6. Generative Adversarial Networks (GANs)

7. Temporal Dynamics

8. Artificial Intelligence
 CHAPTER 3
INTRODUCTION
3.1 Project Idea
The project aims to develop a deepfake video detection system using a hybrid deep
learning approach. With the rise of deepfake technology, it has become challenging to
distinguish real from fake content, posing risks in security, media, and personal privacy.
Our solution combines Convolutional Neural Networks (CNN) for spatial feature extrac-
tion and Recurrent Neural Networks (RNN) with Long Short-Term Memory (LSTM) for
temporal analysis. Additionally, Generative Adversarial Network (GAN) verification is
used to enhance accuracy. Using FFmpeg and OpenCV, frames are extracted and pro-
cessed from videos to train the model, which is then tested and evaluated using metrics
like accuracy and confusion matrix. This project aims to create a robust, automated
tool to detect deepfakes, providing a reliable method for content verification in real-world
applications.

3.2 Motivation of the Project

The widespread use of deepfake technology has made it increasingly difficult to trust
visual media.Deepfakes, which use AI to create realistic but fake videos, can be misused
for malicious purposes, including spreading misinformation, impersonating individuals,
and damaging reputations. This presents a significant threat to personal privacy, social
trust, and public security, as manipulated content can influence public opinion, deceive
viewers, or exploit individuals.

Given the sophistication of modern deepfake generation techniques, traditional meth-

ods for identifying altered content are often insufficient. Automated, accurate detection
methods are essential to counter the risks posed by deepfakes. This project is motivated
by the need to develop a reliable, effective deepfake detection system that leverages ad-
vanced deep learning models (such as CNNs, RNNs/LSTMs, and GAN verification) to
distinguish between real and manipulated videos. Our goal is to provide a tool that en-
hances digital media integrity and helps safeguard against the misuse of deepfake technol-
ogy in various fields, including media verification, cybersecurity, and content moderation.

3.3 Literature Survey

Paper Name Findings Dataset Used Algorithm Accuracy
Used Rate

DFFMD: A Introduced DFFMD Inception- 99.81%

Deepfake Face Deepfake Face Dataset ResNet-
Mask Dataset Mask Dataset (masked v2, Batch
for Infectious (DFFMD) videos) Normalization
Disease Era to improve
with Deep- detection in
fake Detection masked videos.
Algorithms Achieved
99.81% ac-
curacy using
Inception-
ResNet-v2
with batch
normalization
and residual
connections.

Improving Proposed us- Celeb-DF v2 Video Vision 87.18%

Video Vision ing ViViT Transformer
Transformer for deepfake (ViViT),
for Deep- detection fo- Depthwise
fake Video cusing on facial Separable
Detection landmarks. Convolution
Using Facial Achieved (DSC), CBAM
Landmark, 87.18% ac-
Depthwise curacy on
Separable Con- Celeb-DF
volution, and v2 dataset
Self-Attention by utilizing
Depthwise
Separable Con-
volution and
CBAM.

Reviewed Multiple CNNs, RNNs, Varies by

DeepFake deepfake detec- datasets Optical Flow, method (e.g.,
Detection for tion methods (e.g., Face- XceptionNet 85-95% for
Human Face such as CNNs Forensics++, XceptionNet)
Images and for pixel anal- DFDC)
Videos: A ysis and RNNs
Survey for temporal
consistency.
Emphasized
challenges in
generalization
and dataset
limitations.

SIEM, Department of Computer Engineering 2019-20 10

 CHAPTER 4
PROBLEM DEFINITION AND SCOPE
4.1 Problem Statement
With the rise of deepfake technology, there is an crucial need for effective detection meth-
ods to prevent the misuse of manipulated media in security, privacy, and information
integrity contexts. Current detection models struggle to generalize across varied deep-
fake types, and there is a need for approaches that leverage both spatial and temporal
features for improved accuracy. This project aims to develop a robust deepfake detection
system using a hybrid deep learning approach involving Convolutional Neural Networks
(CNN), Long Short-Term Memory (LSTM) networks, and Generative Adversarial Net-
work (GAN) verification. By leveraging FFmpeg for frame extraction and OpenCV for
spatial feature analysis, the model will be trained and tested on preprocessed datasets
split in a 70-30 format. The system’s effectiveness will be evaluated using a confusion
matrix to ensure reliable and accurate detection of deepfake content.

4.1.1 Goals and objectives

Goal and Objectives:

• To develop an efficient and accurate deepfake detection model.

• To evaluate the system on challenging datasets and compare its performance with
existing methods.

• To analyze and compare the performance of different deep learning architectures

for deepfake detection.

• To implement a user-friendly interface for real-time detection of deepfake videos.

4.1.2 Statement of scope

This project aims to develop a deepfake detection system that addresses the threats posed
by manipulated media to security, privacy, and information integrity. It processes video
inputs in formats like MP4, AVI, and MKV, with a maximum size and a minimum dura-
tion of 5 seconds. Utilizing a hybrid deep learning approach that integrates Convolutional
Neural Networks (CNN), Long Short-Term Memory (LSTM) networks, and Generative
Adversarial Networks (GAN) for verification, the system leverages FFmpeg for frame
extraction and OpenCV for feature analysis. Input validation ensures supported formats
and acceptable parameters. The outputs include detection results (real or deepfake),
performance metrics, and a confusion matrix, enhancing the accuracy and reliability of
deepfake detection

4.2 Major Constraints

• Data Availability and Quality: The effectiveness of your model heavily depends
on the availability of large, high-quality datasets of both real and deepfake videos.
Datasets need to cover a wide range of face-swap techniques, lighting conditions,
and facial expressions to ensure the model generalizes well.

SIEM, Department of Computer Engineering 2019-20 12

 • False Positives and Negatives: Balancing the model’s sensitivity is crucial. A
model that is too sensitive could result in many false positives (classifying real
videos as fake), while a model that is too lenient might miss false negatives (failing
to detect fake videos).

• Long Training Time: CGiven the complexity of the hybrid model (CNN, RNN,
GAN), training can take a considerable amount of time, especially when work-
ing with large datasets. Long training times can slow down model iterations and
improvements

4.3 Methodologies of Problem solving and efficiency

issues
While there are numerous advanced tools available for producing Deepfake (DF) con-
tent, there aren’t many effective solutions for identifying it. We firmly believe that our
innovative method of DF detection will significantly aid in preventing the spread of mis-
leading and harmful Deepfake videos on the internet.

The figure 4.1 illustrates the workflow of a deepfake detection project, starting with the
dataset (DDFMD), which includes both real and deepfake videos. In the processing stage,
the videos are split into individual frames, where face detection is performed to identify
facial regions. These detected faces are then cropped to focus on the relevant portions of
the frames, and the cropped videos are saved to form the processed dataset. This dataset
is subsequently split into training and testing subsets for model development.

A data loader manages loading these subsets in batches to optimize memory usage
during training. The deepfake detection model consists of a CNN for extracting spatial
features from the frames, an LSTM for learning temporal relationships between frames,
and a ResNext architecture for advanced feature extraction to enhance detection capabil-
ity. For model evaluation, a confusion matrix is used to analyze the model’s performance
in terms of true and false predictions. Finally, the trained model is exported for practical
use, while the processed dataset and results can be used iteratively for further model
refinement.

SIEM, Department of Computer Engineering 2019-20 13

 Figure 4.1: System Architecture

4.4 Outcome
The expected outcome is a highly accurate and robust DeepFake video detection system
that can identify manipulated videos in real-time. This system will significantly enhance
security and privacy by mitigating the risks associated with deceptive content. It will
offer a user-friendly interface, be scalable for various data loads, and integrate seamlessly
with existing platforms. Designed for continuous improvement, the system will adapt
to evolving DeepFake technologies and provide detailed reporting to support effective
decision-making and response.

4.5 Applications
• Social Media Moderation: Detects and removes deepfake videos to prevent
misinformation, protecting users from manipulated content.

• News Verification: Ensures authenticity of video content before publication,

maintaining journalistic integrity.

SIEM, Department of Computer Engineering 2019-20 14

 • Legal Evidence Analysis: Validates video evidence in legal cases, deepfakes from
compromising investigations.

• Corporate Security: Protects businesses from impersonation in virtual meetings,

interviews, and digital communications.

• Government Security: Monitors for deepfake threats in national security and

intelligence, helping to identify malicious propaganda or influence operations.

4.6 Hardware Resources Required

Sr. No. Parameter Minimum Requirement Justification

1 CPU Speed 2 GHz Remark Required
2 RAM 3 GB Remark Required

Table 4.1: Hardware Requirements

4.7 Software Resources Required

Platform :

1. Operating System:

2. IDE:

3. Programming Language
 CHAPTER 5
PROJECT PLAN
5.0.1 Reconciled Estimates
This section provides a detailed, unified projection of the estimated costs and time
required to complete the project, adjusted to fit the constraints of a student budget
and schedule. This reconciled estimate consolidates various cost estimates, taking into
account available resources and optimizing for affordability and efficiency.

5.0.1.1 Cost Estimate

The following table provides a breakdown of the estimated expenses in Indian Rupees
(INR) for developing a deepfake detection system using a hybrid model (CNN, RNN,
GAN verification).

Item Description Estimated Cost (INR)

Hardware Utilizing a high- 0
performance laptop or
PC with GPU (for model
training). (Assuming
existing system)
Cloud Services Cloud GPU for model 7,000
training and testing (e.g.,
AWS, Google Cloud) - 50
hours of usage
Software and Tools Licenses for specialized 0
software tools, if any (most
libraries are free for stu-
dents) .
Dataset Acquisition Access to datasets (Face- 0
Forensics++, DFDC) -
usually free for academic
purposes
Internet Expenses High-speed internet for 1,500
data transfer and model
training (for 3 months)
Utilities Electricity costs for ad- 1,500
ditional system usage
due to heavy processing
(estimated)
Miscellaneous Printing, binding, and 2,500
other project report prepa-
ration costs
Learning Resources Online courses or books 2,000
for improving skills related
to deepfake detection (if
required)

Total Estimated Cost: 14,500

SIEM, Department of Computer Engineering 2019-20 17

5.0.1.2 Time Estimates

Figure 5.1: Timeline Chart

5.0.2 Project Resources

Project resources [People, Hardware, Software, Tools and other resources] based on Mem-
ory Sharing, IPC, and Concurrency derived using annexures to be referred.

5.1 Risk Management w.r.t. NP Hard analysis

In the context of hybrid deepfake detection model (combining CNN, RNN, and GAN
verification), the system involves complex computational challenges. NP-Hard prob-
lems often arise in optimization, decision-making, and real-time processing. Below is a
risk management strategy based on NP-Hard analysis, covering Risk Identification, Risk
Analysis, and Risk Mitigation, Monitoring, and Management.

5.1.1 Risk Identification

For risks identification, review of scope document, requirements specifications and sched-
ule is done. Answers to questionnaire revealed some risks. Each risk is categorized as
per the categories mentioned. Please refer table 5.1 for all the risks. You can refereed
following risk identification questionnaire.

1. Will the computational complexity of the model lead to performance bottlenecks

in real-time detection?
Yes

SIEM, Department of Computer Engineering 2019-20 18

 2. Is it possible that the hybrid model will require more computing resources than
available hardware can support?
Yes
3. Can GAN verification result in unstable model training and inconsistent results?
Yes
4. Will handling live video streams pose a significant risk of latency or delays?
Yes
5. Is there a chance that the model will struggle to generalize to new and evolving
deepfake techniques?
Yes
6. Is the risk of encountering false positives in deepfake detection low with this model?
No
7. Will using cloud computing mitigate all performance and resource issues?
No
8. Is there a chance that the model will fail to detect subtle deepfake manipulations
due to limited training data?
9. Can real-time audio-visual synchronization detection be affected by poor-quality
video or noisy audio?
Yes

5.1.2 Risk Analysis

The risks for the Project can be analyzed within the constraints of time and quality

Impact
ID Risk Description Probability
Schedule Quality Overall
1 Computational complexity leads High High Medium High
to real-time detection bottlenecks
2 Resource constraints exceed Medium High Medium High
hardware capabilities
3 GAN verification instability caus- Medium-High Medium High High
ing model inconsistency
4 Latency in live video stream Medium-High High Medium High
processing

Table 5.1: Risk Table

5.1.3 Overview of Risk Mitigation, Monitoring, Management

Following are the details for each risk.

SIEM, Department of Computer Engineering 2019-20 19

Probability Value Description
Low If there are ample resources The risk may happen in specific conditions
and needs to be monitored closely.
Medium If some data is outdated There is a moderate chance of this risk oc-
curring (20% - 50%).
Medium-High If partial hardware failure occurs This risk is likely to occur in certain circum-
stances (50% - 75%).
High If network resources are limited There is a high probability that this risk will
occur (¿ 75%).

Table 5.2: Risk Probability Definitions

Impact Value Description

Very high > 10% If a critical error occurs, leading to significant project delays
or failure.
High 6% − 10% If major issues arise, causing noticeable delays or quality
reduction.
Medium 3% − 5% If moderate issues occur, affecting the schedule or quality but
manageable.
Low < 3% If minor issues happen, with minimal impact on the schedule
or quality.

Table 5.3: Risk Impact Definitions

SIEM, Department of Computer Engineering 2019-20 20

Risk Description Impact on Development Envi-
ronment
Limited Network Bandwidth Slower data transfers, affecting model
training and real-time testing.
Insufficient Hardware Resources (CPU/GPU) Slows down model training and in-
creases overall development time.
Tool Compatibility Issue Delays due to incompatibility between
development tools and frameworks.
Software Bugs and Crashes Increased time spent debugging and
fixing crashes, delaying development.
Version Control Conflicts Collaboration issues, potential loss of
code, and increased merge conflicts.
Outdated Libraries and Frameworks Decreased performance, compatibility
problems with modern systems.
Security Vulnerabilities in Development Tools Potential security risks such as
data breaches or compromised code
integrity.

Table 5.4: Risk for Development Environment

Risk ID 2
Risk Description Description 2
Category Requirements
Source Software Design Specification documentation review.
Probability Low
Impact High
Response Mitigate
Strategy Better testing will resolve this issue.
Risk Status Identified

Table 5.5: Risk for Requirements

SIEM, Department of Computer Engineering 2019-20 21

 Risk ID 3
Risk Description Description 3
Category Technology
Source This was identified during early development and testing.
Probability Low
Impact Very High
Response Accept
Strategy Example Running Service Registry behind proxy balancer
Risk Status Identified

Table 5.6: Risk for Technology

5.2 Project Schedule

5.2.1 Project task set
Major Tasks in the Project stages are:

5.2.2 Task network

Project tasks and their dependencies are noted in this diagrammatic form.

5.2.3 Timeline Chart

A project timeline chart is presented. This may include a time line for the entire project.
Above points should be covered in Project Planner as Annex C and you can mention here
Please refer Annex C for the planner

5.3 Team Organization

The manner in which staff is organized and the mechanisms for reporting are noted.

5.3.1 Team structure

The team structure for the project is identified. Roles are defined.

5.3.2 Management reporting and communication

Mechanisms for progress reporting and inter/intra team communication are identified as
per assessment sheet and lab time table.
 CHAPTER 6
SOFTWARE REQUIREMENT
SPECIFICATION
6.1 Introduction
6.1.1 Purpose and Scope of Document
The purpose of SRS and what it covers is to be stated

6.1.2 Overview of responsibilities of Developer

What all activities carried out by developer?

6.2 Usage Scenario

This section provides various usage scenarios for the system to be developed.

6.2.1 User profiles

The profiles of all user categories are described here.(Actors and their Description)

6.2.2 Use-cases
All use-cases for the software are presented. Description of all main Use cases using use
case template is to be provided.

Sr No. Use Case Description Actors Assumptions

1 Use Case 1 Description Actors Assumption

Table 6.1: Use Cases

6.2.3 Use Case View

Use Case Diagram. Example is given below

6.3 Data Model and Description

6.3.1 Data Description
Data objects that will be managed/manipulated by the software are described in this
section. The database entities or files or data structures required to be described. For
data objects details can be given as below

6.3.2 Data objects and Relationships

Data objects and their major attributes and relationships among data objects are de-
scribed using an ERD- like form.

SIEM, Department of Computer Engineering 2019-20 24

 Figure 6.1: Use case diagram

SIEM, Department of Computer Engineering 2019-20 25

6.4 Functional Model and Description
A description of each major software function, along with data flow (structured analy-
sis) or class hierarchy (Analysis Class diagram with class description for object oriented
system) is presented.

6.4.1 Data Flow Diagram

6.4.1.1 Level 0 Data Flow Diagram
6.4.1.2 Level 1 Data Flow Diagram

6.4.2 Activity Diagram:

• The Activity diagram represents the steps taken.

6.4.3 Non Functional Requirements:

• Interface Requirements

• Performance Requirements

• Software quality attributes such as availability [ related to Reliability], modifiability

[includes portability, reusability, scalability] , performance, security, testability and
usability[includes self adaptability and user adaptability]

6.4.4 State Diagram:

State Transition Diagram
Fig.6.2 example shows the state transition diagram of Cloud SDK. The states are rep-
resented in ovals and state of system gets changed when certain events occur. The
transitions from one state to the other are represented by arrows. The Figure shows
important states and events that occur while creating new project.

6.4.5 Design Constraints

Any design constraints that will impact the subsystem are noted.

6.4.6 Software Interface Description

The software interface(s)to the outside world is(are) described. The requirements for
interfaces to other devices/systems/networks/human are stated.
 Figure 6.2: State transition diagram

SIEM, Department of Computer Engineering 2019-20 27

 CHAPTER 7
DETAILED DESIGN DOCUMENT USING
ANNEXURE A AND B
7.1 Introduction
This document specifies the design that is used to solve the problem of Product.

7.2 Architectural Design

A description of the program architecture is presented. Subsystem design or Block dia-
gram,Package Diagram,Deployment diagram with description is to be presented.

Figure 7.1: Architecture diagram

7.3 Data design (using Annexures A and B)

A description of all data structures including internal, global, and temporary data struc-
tures, database design (tables), file formats.

SIEM, Department of Computer Engineering 2019-20 29

7.3.1 Internal software data structure
Data structures that are passed among components the software are described.

7.3.2 Global data structure

Data structured that are available to major portions of the architecture are described.

7.3.3 Temporary data structure

Files created for interim use are described.

7.3.4 Database description

Database(s) / Files created/used as part of the application is(are) described.

7.4 Compoent Design

Class diagrams, Interaction Diagrams, Algorithms. Description of each component de-
scription required.

7.4.1 Class Diagram

 Figure 7.2: Class Diagram

SIEM, Department of Computer Engineering 2019-20 31

 CHAPTER 8
PROJECT IMPLEMENTATION
8.1 Introduction
8.2 Tools and Technologies Used
8.3 Methodologies/Algorithm Details
8.3.1 Algorithm 1/Pseudo Code
8.3.2 Algorithm 2/Pseudo Code

8.4 Verification and Validation for Acceptance

 CHAPTER 9
CONCLUSION AND FUTURE SCOPE
Write summary , conclusion and future scope
REFERENCES
CHAPTER 10
REFERENCES
[1] Anna Cinzia Squicciarini, Dan Lin, Smitha Sundareswaran “Privacy Policy Inference of User-
Uploaded Images on Content Sharing Sites”, in IEEE TRANSACTIONS ON KNOWLEDGE AND
DATA ENGINEERING, VOL. 27, NO. 1, 2015, pp. 193-206.
[2] J. Zhuang and S. C. H. Hoi, “Non-parametric kernel ranking approach for social image retrieval,”
in Proc. ACM Int. Conf. Image Video Retrieval, 2010, pp. 26–33. [Online]. Available: http://doi.
acm.org/10.1145/1816041.1816047
[3] S. Zerr, S. Siersdorfer, J. Hare, and E. Demidova, “Privacy-aware image classification and search,”
in Proc. 35th Int. ACM SIGIR Conf. Res. Develop. Inform. Retrieval, 2012, pp. 35–44.
[4] C.-H. Yeh, Y.-C. Ho, B. A. Barsky, and M. Ouhyoung, “Personalized photograph ranking and selec-
tion system,” in Proc. Int. Conf. Multimedia, 2010, pp. 211–220. Available: http://doi.acm.org/
10.1145/1873951.1873963
[5] K. Strater and H. Lipford, “Strategies and struggles with privacy in an online social networking
community,” in Proc. Brit. Comput. Soc. Conf. Human-Comput. Interact., 2008, pp.111–119
[6] A. Acquisti and R. Gross, “Imagined communities: Awareness,information sharing, and privacy on
the facebook,” in Proc. 6th Int. Conf. Privacy Enhancing Technol. Workshop, 2006,pp. 36–58.
[7] R. Agrawal and R. Srikant,“Fast algorithms for mining association rules in large databases,” in Proc.
20th Int. Conf. Very Large Data Bases, 1994, pp. 487–499.
[8] S. Ahern, D. Eckles, N. S. Good, S. King, M. Naaman, and R. Nair, “Over-exposed?: Privacy
patterns and considerations in online and mobile photo sharing,” in Proc. Conf. Human Factors
Comput. Syst., 2007, pp. 357–366.
[9] S. Ahern, D. Eckles, N. S. Good, S. King, M. Naaman, and R. Nair, “Over-exposed?: Privacy
patterns and considerations in online and mobile photo sharing,” in Proc. Conf. Human Factors
Comput. Syst., 2007, pp. 357–366.
[10] M. Ames and M. Naaman, “Why we tag: Motivations for annotation in mobile and online media,”
in Proc. Conf. Human Factors Comput. Syst., 2007, pp. 971–980.
[11] A. Besmer and H. Lipford, “Tagged photos: Concerns, perceptions, and protections,” in Proc. 27th
Int. Conf. Extended Abstracts Human Factors Comput. Syst., 2009, pp. 4585–4590.
[12] S. B. Barnes. A privacy paradox: Social networking in the united states. First Monday, 11(9), Sept.
2006.
[13] J. Bonneau, J. Anderson, and G. Danezis, “Prying data out of a social network,” in Proc. Int. Conf.
Adv. Soc. Netw. Anal. Mining., 2009, pp.249–254.
[14] H.-M. Chen, M.-H. Chang, P.-C. Chang, M.-C. Tien, W. H. Hsu,and J.-L. Wu, “Sheepdog: Group
and tag recommendation for flickr photos by automatic search-based learning,” in Proc. 16th ACM
Int. Conf. Multimedia, 2008, pp. 737–740.
[15] M. D. Choudhury, H. Sundaram, Y.-R. Lin, A. John, and D. D. Seligmann, “Connecting content
to community in social media via image content, user tags and user communication,” in Proc. IEEE
Int. Conf. Multimedia Expo, 2009, pp.1238–1241.
[16] V. Schleswig-Holstein. Statistische erfassung zum internetverhalten jugendlicher und heranwach-
sender. In A study of the consumer organization in Schleswig-Holstein, Germany, March 2010.
[17] R. da Silva Torres and A. Falc ao, “Content-based image retrieval: Theory and applications,”
Revista de Informatica Teorica e Aplicada, vol. 2, no. 13, pp. 161–185, 2006.
[18] J. Deng, A. C. Berg, K. Li, and L. Fei-Fei, “What does classifying more than 10,000 image categories
tell us?” in Proc. 11th Eur. Conf. Comput. Vis.: Part V, 2010, pp. 71–84. [Online]. Available:
http://portal.acm.org/citation.cfm?id=1888150.1888157
[19] A. Kapadia, F. Adu-Oppong, C. K. Gardiner, and P. P. Tsang, “Social circles: Tackling privacy in
social networks,” in Proc. Symp. Usable Privacy Security, 2008
 ANNEXURE A
LABORATORY ASSIGNMENTS ON
PROJECT ANALYSIS OF
ALGORITHMIC DESIGN
• To develop the problem under consideration and justify feasibilty using concepts of knowledge
canvas and IDEA Matrix.
Refer [?] for IDEA Matrix and Knowledge canvas model. Case studies are given in this book.
IDEA Matrix is represented in the following form. Knowledge canvas represents about identifica-
tion of opportunity for product. Feasibility is represented w.r.t. business perspective.

I D E A
Increase Drive Educate Accelerate
Improve Deliver Evaluate Associate
Ignore Decrease Eliminate Avoid
Table A.1: IDEA Matrix

• Project problem statement feasibility assessment using NP-Hard, NP-Complete or satisfy ability
issues using modern algebra and/or relevant mathematical models.
• input x,output y, y=f(x)
 ANNEXURE B
LABORATORY ASSIGNMENTS ON
PROJECT QUALITY AND RELIABILITY
TESTING OF PROJECT DESIGN
It should include assignments such as
• Use of divide and conquer strategies to exploit distributed/parallel/concurrent processing of the
above to identify object, morphisms, overloading in functions (if any), and functional relations
and any other dependencies (as per requirements). It can include Venn diagram, state diagram,
function relations, i/o relations; use this to derive objects, morphism, overloading
• Use of above to draw functional dependency graphs and relevant Software modeling methods,
techniques including UML diagrams or other necessities using appropriate tools.
• Testing of project problem statement using generated test data (using mathematical models, GUI,
Function testing principles, if any) selection and appropriate use of testing tools, testing of UML
diagram’s reliability. Write also test cases [Black box testing] for each identified functions. You
can use Mathematica or equivalent open source tool for generating test data.
• Additional assignments by the guide. If project type as Entreprenaur, Refer [?],[?],[?], [?]
 ANNEXURE C
PROJECT PLANNER
Using planner or alike project management tool.
 ANNEXURE D
REVIEWERS COMMENTS OF PAPER
SUBMITTED
(At-least one technical paper must be submitted in Term-I on the project design in the conferences/workshops
in IITs, Central Universities or UoP Conferences or equivalent International Conferences Sponsored by
IEEE/ACM)
1. Paper Title:
2. Name of the Conference/Journal where paper submitted :
3. Paper accepted/rejected :
4. Review comments by reviewer :
5. Corrective actions if any :
 ANNEXURE E
PLAGIARISM REPORT
Plagiarism report
 ANNEXURE F
INFORMATION OF PROJECT GROUP
MEMBERS
one page for each student .

SIEM, Department of Computer Engineering 2019-20 50

1. Name :
2. Date of Birth :
3. Gender :
4. Permanent Address :
5. E-Mail :
6. Mobile/Contact No. :
7. Placement Details :
8. Paper Published :

SIEM, Department of Computer Engineering 2019-20 51