FAKE IMAGE IDENTIFICATION USING

CONVOLUTIONAL NEURAL NETWORK (CNN)

An industry oriented Major Project Report Submitted to

Jawaharlal Nehru Technological University Hyderabad

In partial fulfillment of the requirements for the award of the

degree of
BACHELOR OF TECHNOLOGY
IN
COMPUTER SCIENCE AND ENGINEERING
Submitted By
M.GUNA SIDDARTH (21E11A052O)
A. ANIL KUMAR (21E11A0502)
A. AJAY KUMAR (22E11A050)
S.VAMSHI (21E11A0531)

Under the Supervision of

MR.G.RAGHAVENDER
Associate professor, CSE department.

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

BHARAT INSTITUTE OF ENGINEERING AND TECHNOLOGY
Accredited by NAAC, Accredited by NBA (UG Programmes: CSE, ECE)
Approved by AICTE, Affiliated to JNTUH Hyderabad
Hyderabad-501 510, Telangana. Jan 2025

i
 DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING BHARAT
INSTITUTE OF ENGINEERING AND TECHNOLOGY
Accredited by NAAC, Accredited by NBA (UG Programmes: CSE, ECE)
Approved by AICTE, Affiliated to JNTUH Hyderabad
Hyderabad-501 510, Telangana.

Certificate
This is to certify that the project work entitled “Sign Language Translation Using
Convolutional Neural Networks (CNN)” is the bonafide work done

By
M.GUNA SIDDARTH (21E11A0520)
A.ANIL KUMAR (21E11A0502)
A.AJAY KUMAR (22E11A0501)
S.VAMSHI (21E11A0531)

in the Department of Computer Science and Engineering, BHARAT INSTITUTE OF

ENGINEERING AND TECHNOLOGY, Hyderabad is submitted to Jawaharlal Nehru
Technological University, Hyderabad in partial fulfillment of the requirements for the award of B. Tech
degree in Computer Science and Engineering during 2024-2025.
Guide: Head of the Department:
Mr. Raghavender Dr. Deepak Kachave
ASSISTANT PROFESSOR ASSOCIATE PROFESSOR
Dept of CSE, Dept of CSE,
Bharat Institute of Engineering and Bharat Institute of Engineering and
Technology, Technology,
Hyderabad – 501 510 Hyderabad – 501 510

Viva-Voce held on……………………………………………

ii
 Internal Examiner External Examiner
ACKNOWLEDGEMENT
The satisfaction that accompanies the successful completion of the task would be put incomplete
without the mention of the people who made it possible, whose constant guidance and encouragement
crown all the efforts with success.

We avail this opportunity to express our deep sense of gratitude and hearty thanks to Shri CH.
Venugopal Reddy, Secretary & Correspondent of BIET, for providing congenial atmosphere and
encouragement.

We would like to thank Prof. G. Kumaraswamy Rao, Director, Former Director & O.S. of DLRL
Ministry of Defence, and Dr. V. Srinivas Rao, Dean CSE for having provided all the facilities and
support.

We would like to thank our Academic Incharge Dr. Deepak Kachave, Associate Professor of CSE, for
their expert guidance and encouragement at various levels of Project.

We are thankful to our Project Supervisor Mr.Raghavender, Assistant Professor, Computer

Science and Engineering for his support and cooperation throughout the process of this Project.

We are thankful to Project Coordinator Dr. Rama Prakasha Reddy Ch, Assistant Professor, Computer
Science and Engineering for his support and cooperation throughout the process of this project.

We place highest regards to our Parent, our Friends and Well-wishers who helped a lot in making the
report of this project.

DECLARATION

We hereby declare that this Project Report (Phase-1) is titled “Fake image identification
Using Convolutional Neural Networks (CNN)” is a genuine project work carried out by us, in
B.Tech (Computer Science and Engineering) degree course of Jawaharlal Nehru Technology
University Hyderabad, and has not been submitted to any other course or university for the award of
my degree by us.

Candidate Name(s) Roll Number Signature

M.GUNA SIDDARTH (21E11A0520)

iii
 A.ANIL KUMAR (21E11A0502)
A.AJAY KUMAR
(22E11A0501) S.VAMSHI
(21E11A0531)

DATE:

ABSTRACT
The rapid advancement of digital imaging technologies has made it easier to create and
manipulate images, which has led to an increase in the distribution of fake or
manipulated images across various platforms. These fake images can have serious
implications in fields such as journalism, social media, and security. In this context,
detecting fake images has become a critical task. This paper proposes a Convolutional
Neural Network (CNN)-based approach for identifying fake images. CNNs, known for
their effectiveness in image recognition tasks, are employed to learn deep hierarchical
features from authentic and fake image datasets. The proposed method involves
preprocessing the images to extract relevant features, followed by training the CNN
model to distinguish between real and fake images based on these features. The system is
evaluated on various benchmark datasets, achieving high accuracy in fake image
detection. Additionally, the model's ability to generalize across different types of fake
images, including those manipulated using techniques like splicing and copy-move, is
discussed. The results demonstrate that the CNN-based approach is a powerful tool for
detecting fake images, offering a promising solution for countering the growing problem
of image forgery.

Keywords: neural networks.

iv
 TABLE OF CONTENTS

S. No Contents Page no

Cover Page i
Certificate ii

Acknowledgement iii
Declaration iv
Abstract v

Table of content vii

List of Figures viii

List of Symbols and Abbreviations ix

v
1 INTRODUCTION

1.1 Introduction to CNN ............................................................................ 14

1.2 Architecture of CNN ........................................................................... 15
2 RELATED WORK ....................................................................................... 19
3 MOTIVATION .............................................................................................. 23
4 OBJECTIVES ................................................................................................ 24
5 PROBLEM STATEMENT
6 SYSTEM DESIGN
6.1 System Requirements Specifications .................................................. 28
6.2 Algorithm .......................................................................................... 52
6.3 Data Flow Diagrams ........................................................................... 54
6.4 UML Diagrams ................................................................................... 56
7 CONCLUSION AND FUTURE SCOPE ................................................... 78
8 REFERENCES ............................................................................................. 79
LIST OF FIGURES
Fig No Title Page No

1.1 Structure of convolutional Neural Network 10

1.2 Internal Structure of CNN with corresponding Layers 11
1.3 Average pooling with corresponding Layers 13
1.4 Fully Connected Layers 14

Dataflow diagram 23

UML Diagram 24

vi
vii
B.Tech – Project phase-1 2024-2025

LIST OF SYMBOLS AND ABBREVIATIONS

Sl.no SYMBOL ABBREVIATION

1 CNN Convolutional Neural Network

2 AI Artificial Intelligence

3 ML Machine Learning

4 IDE Integrated Development Environment

5 IP Internet Protocol

6 SRS Software Requirement Specification

7 SVM Support Vector Machine

8 UML Unified Modelling Language

9 DFD Data Flow Diagram

B.Tech – Project phase-1 2024-2025

viii

1. INTRODUCTION

1.1 INTRODUCTION TO CNN

Convolutional Neural Networks (CNNs) are a class of deep learning models that have
revolutionized the field of computer vision. Inspired by the structure and function of the
human visual system, CNNs are designed to automatically and adaptively learn spatial
hierarchies of features from input images. Unlike traditional machine learning methods
that rely heavily on manual feature extraction, CNNs can learn relevant features directly
from the raw image data during the training process, making them particularly well-suited
for tasks such as image classification, object detection, and image segmentation. regular
Neural Networks, in the layers of CNN, the neurons are arranged
in 3 dimensions: width, height, depth.

9|Page
B.Tech – Project phase-1 2024-2025

Fig 1.1 Structure of convolutional Neural Network

1.2 The architecture of an convolutional neural network:

The Convolutional layer applies filters to the input image to extract features, the
Pooling layer down samples the image to reduce computation, and the fully
connected layer makes the final prediction. The network learns the optimal filters
through backpropagation and gradient descent.

Convolutional Neural Network primarily consists layers:

10 | P a g e
B.Tech – Project phase-1 2024-2025

Fig 1.2 Internal Structure of CNN with corresponding Layers

1. Convolutional Layer:
In convolution layer I have taken a small window size [typically of length 5*5] that extends to
the depth of the input matrix.
The layer consists of learnable filters of window size. During every iteration I slid the window
by stride size [typically 1], and compute the dot product of filter entries and input values at a
given position.
As I continue this process well create a 2-Dimensional activation matrix that gives the
response of that matrix at every spatial position.
That is, the network will learn filters that activate when they see some type of visual feature
such as an edge of some orientation or a blotch of some colour.

2. Pooling Layer:
We use pooling layer to decrease the size of activation matrix and ultimately reduce the
learnable parameters.
There are two types of pooling:

a. Max Pooling:

11 | P a g e
B.Tech – Project phase-1 2024-2025
In max pooling we take a window size [for example window of size 2*2], and only taken the
maximum of 4 values.
Well lid this window and continue this process, so well finally get an activation matrix half of
its original Size.

b. Average Pooling:
In average pooling we take average of all Values in a window.

Fig 1.3 Average pooling with corresponding Layers

12 | P a g e
B.Tech – Project phase-1 2024-2025
3. Fully Connected Layer:
In convolution layer neurons are connected only to a local region, while in a fully connected
region, well connect the all the inputs to neurons.

Fig 1.4 Fully Connected Layers

4. Final Output Layer:

CSE Department, BIET 13 |P a g e

B.Tech – Project phase-1 2024-2025

2. RELATED WORK

2.1 Literature Review:

TITLE: Early Lexical Approach

AUTHOR: Melichar (1991)

DESCRIPTION: One of the early attempts at computational sign language translation was
through the development of sign lexicons. Melichar's work involved building a
dictionary that mapped hand shapes in sign language to corresponding words in spoken
languages. The goal was to create a static mapping that could be used to translate from
signs to text, focusing on one-to-one mappings.

TITLE: The E-Talk System

AUTHOR: Nehls et al. (1995)

DESCRIPTION: The E-Talk system, developed by Nehls and colleagues in the 1990s, was
an early rule-based approach aimed at translating American Sign Language (ASL) to
written English. It used a set of predefined linguistic rules to convert signs into text.
Despite its limitations, this system laid the groundwork for understanding the need for
complex linguistic rules in sign language translation.

TITLE: A Sign Language Translation System

CSE Department, BIET 14 |P a g e
B.Tech – Project phase-1 2024-2025
AUTHOR: Braffort et al. (2004)

DESCRIPTION: This work explored an early machine translation approach for translating

French Sign Language (LSF) to French text. It involved the creation of a rule-based
system where sign language gestures were mapped to linguistic structures in French.
This research contributed to the development of frameworks that attempted to integrate
both visual gestures and text, setting the stage for later research in multimodal
translation systems.

TITLE: Gesture Recognition Using Image Processing

AUTHOR: Tkach et al. (2010)

DESCRIPTION: This study moved from purely rule-based systems towards image

processing and gesture recognition. Tkach and colleagues focused on the problem of
identifying specific hand gestures and mapping them to corresponding spoken words or
symbols. Their system used simple image processing techniques to track hand shapes and
gestures in static images. Tkach's work was an early foray into using computer vision for
sign language recognition, which became a major focus of future research.

TITLE: Machine Learning for Gesture Recognition

AUTHOR: Lennon et al. (2003)

DESCRIPTION: Lennon and colleagues developed a machine learning-based approach to

recognizing sign language gestures. The team explored various methods of using visual
sensors to track hand and arm movements and convert these into textual or speech

CSE Department, BIET 15 |P a g e

B.Tech – Project phase-1 2024-2025
outputs. This early work helped establish the foundation for later research in deep
learning models and the use of large datasets for training gesture recognition systems.

TITLE: Using Video for Sign Language Translation

AUTHOR: Bridges et al. (1997)

DESCRIPTION: The project by Bridges and colleagues was one of the first to explore using

video as input for sign language recognition. This system aimed to recognize ASL signs
and translate them into English text by analyzing video frames. This research played a
pivotal role in demonstrating the potential of video-based translation systems, which later
evolved into more sophisticated machine learning-based solutions.

TITLE: A Manual Encoding System

AUTHOR: SignWriter (1980s)

DESCRIPTION: While not a translation system in itself, SignWriter, developed in the 1980s

by William Stokoe and others, was an early attempt at creating a standardized written
representation of sign languages. SignWriter was a system to visually represent signs using
a written notation that could be used to transcribe signed languages. SignWriter helped
establish the importance of formalizing sign language grammar and contributed to a better
understanding of how sign languages could be documented and analyzed computationally.

CSE Department, BIET 16 |P a g e

B.Tech – Project phase-1 2024-2025

3. MOTIVATION

The motivation for using Convolutional Neural Networks (CNNs) in this field arises from
their proven ability to analyze visual data effectively. CNNs can detect subtle
inconsistencies and patterns, such as pixel-level anomalies, texture distortions, or lighting
mismatches, that are often imperceptible to the human eye. Unlike traditional image
processing techniques, CNNs can automatically learn and extract these features directly
from data, making them highly suitable for complex and evolving tasks like fake image
detection.
Moreover, the emergence of advanced image generation techniques, such as Generative
Adversarial Networks (GANs), has increased the sophistication of fake content. This has
created a pressing need for equally sophisticated detection methods. CNNs, with their
scalability and adaptability, offer a powerful solution to counter these threats. Their
application extends to critical areas such as media authentication, forensic analysis, online
content moderation, and combating the spread of deepfake videos in political, social, and
financial contexts.
In summary, the motivation for fake image identification using CNNs lies in addressing the
growing challenges posed by digital image manipulation, ensuring the authenticity of visual
content, and leveraging cutting-edge technology to safeguard.

CSE Department, BIET 17 |P a g e

 B.Tech – Project phase-1 2024-2025

4. OBJECTIVES

• The primary objectives of using Convolutional Neural Networks (CNNs) for fake image identification are
centered around ensuring the accuracy, robustness, and scalability of detection systems to address the
challenges posed by manipulated content. Below are the key objectives:
• Accurate Detection of Fake Images:
Develop a CNN-based system capable of identifying fake images with high precision, minimizing false
positives and negatives.
• Automated Feature Extraction:
Leverage the hierarchical feature-learning capability of CNNs to automatically detect inconsistencies and
anomalies, such as texture mismatches, lighting distortions, or edge artifacts, without manual
intervention.
• Generalization Across Manipulation Techniques:
Ensure the model can generalize to a wide range of image manipulation methods, including splicing,
deepfakes, GAN-generated content, and other emerging techniques.
• Improved Robustness and Reliability:
Build a detection system resilient to adversarial attacks and capable of handling diverse datasets,
including unseen manipulations or real-world scenarios.
• Real-Time Detection:
Optimize the computational efficiency of CNN models to enable real-time detection of fake images,
suitable for live applications and streaming content analysis.
• Explainability of Model Decisions:
Incorporate interpretability mechanisms, such as visual saliency maps, to help users understand how and
why a particular image was classified as fake or real.
• Scalability and Deployment Readiness:
Design lightweight CNN models or employ optimization techniques for deployment on resource-
constrained devices, such as smartphones or edge computing platforms.
• Multimodal Integration:
Explore the integration of CNNs with other modalities, such as audio and text, to create a comprehensive
fake content detection framework.
CSE Department, BIET 18 |P a g e
 B.Tech – Project phase-1 2024-2025
•

5. PROBLEM STATEMENT

The widespread availability of advanced image editing tools and generative models, such as
Generative Adversarial Networks (GANs), has significantly increased the prevalence of fake
images. These manipulated images, often indistinguishable from authentic ones, pose a
serious threat to societal trust, digital security, and the integrity of information. Fake images
are being used in various malicious activities, including spreading misinformation, defaming
individuals, and facilitating fraud.
Traditional image processing techniques struggle to keep up with the sophistication of
modern manipulations, as they rely on predefined rules that cannot adapt to novel
manipulation methods. This creates a pressing need for automated, accurate, and scalable
solutions capable of identifying fake images across diverse scenarios.
Convolutional Neural Networks (CNNs) offer a promising approach due to their ability to
automatically extract complex features and detect subtle artifacts present in fake images.
However, existing CNN-based detection systems face challenges in generalizing to unseen
manipulations, achieving real-time performance, and explaining their decisions.
Thus, the problem is to develop a robust, efficient, and interpretable CNN-based system that
can accurately detect fake images across a wide range of manipulation techniques, ensuring
reliability and scalability for real-world applications while addressing ethical considerations
and potential biases.

CSE Department, BIET 19 |P a g e

 B.Tech – Project phase-1 2024-2025

6. SYSTEM DESIGN

6.1 SYSTEM REQUIREMENTS SPECIFICATIONS:

Software Requirements Operating System:

Compatible with Windows, macOS, Android, iOS, or Linux.
Programming Tools and Frameworks:
• Computer Vision: o OpenCV or Mediapipe for gesture and facial
recognition.
• Machine Learning: o TensorFlow, PyTorch, or similar frameworks for
training and deploying AI models.
• Natural Language Processing (NLP): o For translating gestures
into grammatically correct text or speech.

Hardware Requirements
Input Devices:
Camera:
o High-resolution camera for capturing hand gestures and body movements.
o Capable of detecting motion in various lighting conditions.
Processing Hardware:
Computer or Mobile Device:
o A system with sufficient processing power to handle real-time image and video processing.
o Example specifications:
CPU: Quad-core or higher.
RAM: 4GB or more.
GPU (for advanced processing): CUDA-enabled GPU for AI models.

6.2 Python Libraries required in the project:

CSE Department, BIET 20 |P a g e

B.Tech – Project phase-1 2024-2025
6.1.3.1 Python Installation:

3. Libraries:

The most popular libraries which are being used by the developers for the
implementation of machine learning in the existing applications are:

• Python 3.6.6
• TensorFlow 1.11.0
• OpenCV 3.4.3.18
• NumPy 1.15.3
• Matplotlib 3.0.0
• Hun spell 2.0.2
• Keras 2.2.1
• PIL 4.3.0

6.3 Artificial Intelligence application in this project

Machine Learning Methods applied in this project

6.4 ALGORITHMS:

Convolutional Neural Networks (CNNs)

Purpose: Hand Gesture Recognition
• CNNs are ideal for image-based tasks because they excel at extracting spatial
features from images.
• Use Case:
• Recognize static hand gestures (e.g., individual letters or words in sign language).
Popular Architectures:

CSE Department, BIET 21 |P a g e

B.Tech – Project phase-1 2024-2025
VGGNet, ResNet, MobileNet (for lightweight mobile-friendly models).

6.5 DATA FLOW DIAGRAMS relevant to the project:

6.6 UML diagrams related to the project:

CSE Department, BIET 22 |P a g e

B.Tech – Project phase-1 2024-2025

CSE Department, BIET 23 |P a g e

B.Tech – Project phase-1 2024-2025

7.CONCLUSION AND FUTURE SCOPE

CONCLUSION:

The identification of fake images has become increasingly critical in today's digital age,
with the rise of manipulated content like deepfakes and doctored images. Convolutional
Neural Networks (CNNs) have proven to be a powerful tool for this task due to their
capability to extract and learn complex features from images.

FUTURE WORK in Phase 2:

In Phase 2 of the project on fake image identification using CNN, the focus shifts
toward improving model performance, addressing limitations, and expanding the scope
to handle more complex scenarios.

CSE Department, BIET 24 |P a g e

 B.Tech – Project phase-1 2024-2025

8.REFERENCES

"Sign Language Recognition Using Deep Learning"

• This paper discusses the use of Convolutional Neural Networks (CNNs) and Recurrent Neural
Networks (RNNs) for recognizing sign language gestures.
• Source: IEEE Xplore
"A Survey on Sign Language Recognition"
• A comprehensive review of techniques and challenges in sign language recognition.
• Source: Springer Link
"Hand Gesture Recognition Using MediaPipe Framework"
Focuses on using MediaPipe for gesture detection and classification.
Source: arXiv

American Sign Language (ASL) Dataset:

• Contains labeled images/videos of ASL gestures.
• ASL Dataset on Kaggle
RWTH-PHOENIX-Weather Dataset:
• A dataset for continuous sign language recognition.
• Dataset Link
Sign Language MNIST:
• Simplified dataset for ASL alphabets (ideal for beginners).
• Kaggle Lin

CSE Department, BIET 25 |P a g e