TABLE OF CONTENTS

Abstract I
Table of Contents II
List of Figure IV
List of tables VI
Abbreviations VII

i
 ABSTRACT

In the era of social media, digital images serve as a primary medium for information
exchange. However, malicious software poses a significant threat by forging images
to disseminate fake information, necessitating robust digital image forgery detection
techniques. Traditional methods often focus on detecting a specific type of forgery,
such as image splicing or copy-move, limiting their practical applicability in real-
world scenarios where multiple types of forgeries coexist. To address this challenge,
this study explores advanced deep learning architectures for detecting diverse image
forgeries using the CASIA dataset. We evaluate the performance of several state-of-
the-art convolutional neural network (CNN) models, including InceptionV3,
VGG16, VGG19, DenseNet, MobileNetV2, ResNet50, ResNet101, ResNet152,
Xception, and NASNetMobile. These models are benchmarked to identify their
efficacy in detecting forgeries with high accuracy. Additionally, a hybrid model
combining Xception and NASNetMobile is proposed to leverage the strengths of
both architectures, enhancing detection accuracy and robustness. The results
demonstrate the potential of hybrid models in effectively addressing real-world
digital image forgery detection, paving the way for secure information sharing on
digital platforms

i
 TABLE OF CONTENTS

S.NO CONTENT PGNO

1 Chapter 1 1
1.1 Introduction and Objective
1.2 Problem Statement
1.3Implemented models
1.4 Implemented Algorithms
2 Chapter 2 14
2.1 Literature survey
2.2 Feasibility Study
3 Chapter 3 22
3.1 System requirements and specifications
3.1.1 existing system and its disadvantages
3.1.2 proposed system and its advantages
3.2 Functional requirements
3.3 Non-Functional requirements
4 Chapter 4 26
4.1 Methodology
4.2 Hardware requirements
4.3 Software requirements
5 Chapter 5 27
5.1 System design
5.1.1 System architecture
5.2 UML diagrams

ii
6 Chapter 6 37
6.1 System testing
6.2 Test cases
7 Chapter 7 40
7.1 comparison graphs
7.2 Screenshots
8 Chapter 8 48
8.1 Conclusion
8.2 Future scope
References 49

iii
 LIST OF FIGURES

FIG NUM NAME OF FIGURE PAGE NO

Fig 1.1 LAYERS OF SOFTWARE ENVIRONMENT 5

Fig 5.1.1 SYSTEM ARCHITECTURE 28

Fig 5.1.2 DATA FLOW DIAGRAM 29

Fig 5.2.1 UML DAIAGRAM 31

Fig 5.2.2 CLASS DAIAGRAM 32

Fig 5.2.3 ACTIVIVTY DAIGRAM 33

Fig 5.2.4 SEQUENCE DIAGRAM 34

Fig 5.2.5 COLLABORATION DIAGRAM 35

Fig 5.2.6 COMPONENT DIAGRAM 36

Fig 5.2.7 DEPLOYMENT DIAGRAM 36

FIG 6.1.1 TYPES OF SOFTWARE SYSTEM TESTING 38

FIG 7.1.1 ACCURACY COMPARISON GRAPH 41

FIG 7.1.2 PRECISION COMPARISON GRAPH 41

FIG 7.1.3 RECALL COMPARISON GRAPH 42

FIG 7.1.4 F1 COMPARISON GRAPH 43

FIG 7.2.1 INTERFACE OR HOME SCREEN AFTER 43

SIGN IN OR SIGNUP
FIG 7.2.2 44
COLLECTION OF USER INFORMATION
FIG 7.2.3 USER INFORMATION IS COLLECTED 44

iv
 WHILE MAINTAINING CLARITY
AND SIMPLICITY

FIG 7.2.4 45
SIGN IN TO PAGE
FIG 7.2.5 45
UPLOADING OF IMAGE FROM FOLDER
SELECTING REAL PICTURE FROM FOLDER 46
FIG 7.2.6.1
RESULTS SHOWN WILL BE OF REAL 46
FIG 7.2.6.2
IMAGE

FIG 7.2.7.1 47
SELECTING FAKE PICTURE FROM FOLDER
FIG 7.2.7.2 47
RESULTS SHOWN WILL BE OF FAKE IMAGE

v
 LIST OF TABLES

TABLE NO NAME OF THE TABLE PAGE NO

6.2.1 TEST CASES OF THREE DIFFERENT 39
INPUTS
7.1.1 PERFORMANCE EVALUATION TABLE 40

vi
 ABBREVIATIONS

SL NO Abbreviations Full Name

1. CNN Convolutional Neural Networks
2. CASIA Connecticut Alarm and Systems Integrators
Association

3. VGG Visual Geometry Group

4. RNN Residual Neural Network
5. DenseNet Densely Connected Networks
6. NASNet Neural Architecture Search Network
7. GNU Gnu’s Not Unix
8. GINP Gnu Image Manipulation Program
9. RNN Recurrent Neural Networks
10. NLP Natural Language Processing
11. ODTLCMFD Optimal Deep Transfer Learning Based Copy Move
Forgery Detection
12. PO Political Optimizer
13. EBSA Enhanced Bird Swarm Algorithm
14. LS-SVM Least Square Support Vector Machine
15. MSVM Multiclass Support Vector Machine
16. CMFD Copy-Move Forgery Detection
17. ILSVRC ImageNet Large Scale Visual Recognition Challenge
18. SRM Spatial Rich Model
19. SVM Support Vector Machine
20. CRFs Conditional random fields
21. CIAR Canadian Institute for Advanced Research
22. CE Contrast Enhancement
23. MF Median Filtering

vii
DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

CHAPTER 1
INTRODUCTION
1.1. INTRODUCTION
The alteration of a digital picture is referred to as digital image forgery; such manipulated
images are undetectable to the unaided eye. Such photos serve as key sources for
disseminating false information and misleading narratives across society, facilitated by
various social media platforms such as Facebook and Twitter. Free editing software
programs capable of creating these forgeries include powerful functionality for image
manipulation, such as GNU, GIMP, and Adobe Photoshop. Digital picture forgery
algorithms and strategies can identify such forgeries, particularly in image security when
the original material is unavailable. Digital picture counterfeiting refers to the
incorporation of atypical patterns into original photographs, resulting in a heterogeneous
variance in image attributes and an irregular distribution of image components. The
categorization of digital picture manipulation. Active methodologies need critical
information regarding the picture for the verification procedure. The information
embedded in the image is utilized to monitor alterations in that image. The active
technique has two categories: digital signatures, which embed supplementary data derived
from a picture at the conclusion of the acquisition process, and digital watermarking,
which is integrated into images either during the capture phase or the processing phase.

Objective:
The objective of this study is to develop a comprehensive and robust digital image
forgery detection system capable of identifying diverse types of forgeries, such as image
splicing and copy-move, in real-world scenarios. Utilizing the CASIA dataset, the study
evaluates the performance of advanced deep learning architectures, including
InceptionV3, VGG16, VGG19, DenseNet, MobileNetV2, ResNet50, ResNet101,
ResNet152, Xception, and NASNetMobile, to detect subtle inconsistencies in
manipulated images. Additionally, the study proposes a hybrid model combining
Xception and NASNetMobile to leverage their strengths, aiming to enhance detection
accuracy, scalability, and reliability for secure and trustworthy information sharing on
digital platforms.

Department of CSE, GNDEC, Bidar. Page 1

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

1.2 Problem Statement:

 Digital images on social media are susceptible to manipulation by malicious
software, enabling the spread of fake information through techniques like
image splicing and copy-move forgery.
 Traditional forgery detection methods are often limited to specific forgery
types and lack versatility, making them impractical for addressing diverse
forgery techniques found in real-world scenarios.
 Misinformation caused by forged images impacts individuals, communities,
and organizations, leading to reputational damage, financial loss, and
compromised trust in shared digital content.
 The proliferation of forged images undermines the credibility of social media
platforms, fuels the spread of fake news, and complicates efforts to ensure
secure and reliable information dissemination.
 We propose a deep learning-based system using CNN models and a hybrid
Xception-NASNetMobile model to detect diverse forgeries effectively,
ensuring robust and trustworthy digital content verification.

1.3 Implemented Models :

 Data loading: using this module we are going to import the dataset.
 Image processing: Image processing involves using Image Data Generator to
enhance the dataset by performing various transformations. These include re-
scaling the image, applying shear transformation, zooming, horizontal
flipping, and reshaping the image to improve model training and robustness.
 Model generation: Model building - CNN, InceptionV3, VGG16, VGG19,
DenseNet, MobileNetV2, ResNet50, ResNet101, ResNet152, Xception,
NasNetMobile, Hybrid Model - Xception + NASNetMobile.
 User signup & login: Using this module will get registration and login
 User input: Using this module will give input for prediction
 Prediction: final predicted displayed

Department of CSE, GNDEC, Bidar. Page 2

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

Extension:
We can enhance performance by integrating the ensemble of Xception and
NASNetMobile models, improving accuracy and scalability in detecting image forgeries.
Additionally, we can develop a Flask-based frontend for user testing, incorporating
authentication for secure access. This setup allows real-time interaction with the system
and supports seamless deployment for practical use, increasing its utility in real-world
applications.
Advantages:

 Xception model provides superior feature extraction, enhancing forgery

detection accuracy.
 NASNetMobile ensures efficiency and scalability in processing complex image
data.
 Flask framework offers a lightweight, easy-to-deploy interface for user
interaction.
 The Flask framework supports seamless integration with authentication for
enhanced security.
 Real-time testing through the interface makes the system more accessible for
users.

1.4 Implemented Algorithms:

CNN: Utilized for feature extraction and pattern recognition in images, CNN models
effectively identify and classify image forgeries, providing a foundational approach for
detecting subtle image manipulations.
InceptionV3: Known for its efficiency in handling large-scale image data, InceptionV3
leverages deep layers and multi-scale convolutions to detect complex image forgeries
with high precision and low computational cost.
VGG16: A deep convolutional network, VGG16 excels at extracting fine-grained
features from images, enabling the identification of manipulated sections in forgery
detection tasks, ensuring robust classification.

Department of CSE, GNDEC, Bidar. Page 3

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

VGG19: With a deeper architecture than VGG16, VGG19 enhances feature extraction
capabilities, improving the detection of intricate image manipulations and providing a
more detailed analysis of image authenticity.
DenseNet: Dense Net‘s feature reuse across layers facilitates efficient image processing.
It excels at detecting small image forgeries by leveraging its densely connected layers to
capture subtle details in manipulated images.
MobileNetV2: Designed for efficiency, MobileNetV2 processes images faster with fewer
resources, making it ideal for real-time forgery detection without compromising accuracy
in detecting smaller, less obvious manipulations.
ResNet50: ResNet50‘s residual learning allows for deeper networks without performance
degradation, making it effective for forgery detection, especially in handling images with
complex manipulation techniques.
ResNet101: Building upon ResNet50, ResNet101 enhances forgery detection by
providing deeper feature representations, which improves accuracy in detecting subtle
and complex forgeries in digital images.
ResNet152: With even deeper layers than ResNet101, ResNet152 excels at detecting
highly intricate forgeries by offering refined feature extraction and robust handling of
various manipulation techniques.
Xception: Xception utilizes depthwise separable convolutions, allowing for high
efficiency in feature extraction, making it particularly effective in detecting detailed and
sophisticated forgery types with minimal computational overhead.
NASNetMobile: Leveraging neural architecture search, NASNetMobile provides a
lightweight but highly effective model for detecting image forgeries, optimizing accuracy
while maintaining fast processing for mobile or edge devices.
Hybrid Model - Xception + NASNetMobile: Combining Xception‘s powerful feature
extraction and NASNetMobile‘s efficiency, this hybrid model enhances forgery detection
performance, offering both high accuracy and scalability for real- world applications.

1.5 Software Environment Introduction To Deep Learning:

Deep learning is a type of machine learning that uses artificial neural networks to learn
from data. Artificial neural networks are inspired by the human brain, and they can be used
to solve a wide variety of problems, including image recognition, natural language
processing, and speech recognition.

Department of CSE, GNDEC, Bidar. Page 4

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

Figure 1.1 Layers of Software Environment

How does deep learning work?

Deep learning works by using artificial neural networks to learn from data. Neural
networks are made up of layers of interconnected nodes, and each node is responsible for
learning a specific feature of the data. Building on our previous example with images – in
an image recognition network, the first layer of nodes might learn to identify edges, the
second layer might learn to identify shapes, and the third layer might learn to identify
objects.
As the network learns, the weights on the connections between the nodes are adjusted so
that the network can better classify the data. This process is called training, and it can be
done using a variety of techniques, such as supervised learning, unsupervised learning,
and reinforcement learning.
Once a neural network has been trained, it can be used to make predictions with new data
it‘s received.
Deep learning applications:
Deep learning can be used in a wide variety of applications, including:
 Image recognition: To identify objects and features in images, such as people,
animals, places, etc.
 Natural language processing: To help understand the meaning of text, such

Department of CSE, GNDEC, Bidar. Page 5

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

as in customer service chatbots and spam filters.

 Finance: To help analyze financial data and make predictions about market
trends
 Text to image: Convert text into images, such as in the Google Translate app.
 Types of deep learning:
 There are many different types of deep learning models. Some of the most
common types include:
 Convolutional neural networks (CNNs)
 CNNs are used for image recognition and processing. They are particularly
good at identifying objects in images, even when those objects are partially
obscured or distorted.
 Deep reinforcement learning
 Deep reinforcement learning is used for robotics and game playing. It is a type
of machine learning that allows an agent to learn how to behave in an
environment by interacting with it and receiving rewards or punishments.
 Recurrent neural networks (RNNs)
 RNNs are used for natural language processing and speech recognition. They
are particularly good at understanding the context of a sentence or phrase, and
they can be used to generate text or translate languages.
Challenges of using deep learning models:
Deep learning also has a number of challenges, including:
1. Data requirements: Deep learning models require large amounts of data to
learn from, making it difficult to apply deep learning to problems where
there is not a lot of data available.
2. Overfitting: DL models may be prone to overfitting. This means that they
can learn the noise in the data rather than the underlying relationships.
3. Bias: These models can potentially be biased, depending on the data that it‘s
based on. This can lead to unfair or inaccurate predictions. It is important to
take steps to mitigate bias in deep learning models.

What are the uses of deep learning?

Deep learning has several use cases in automotive, aerospace, manufacturing, electronics,
medical research, and other fields. These are some examples of deep learning:

Department of CSE, GNDEC, Bidar. Page 6

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

 Self-driving cars use deep learning models to automatically detect road signs
and pedestrians.
 Defense systems use deep learning to automatically flag areas of interest in
satellite images.
 Medical image analysis uses deep learning to automatically detect cancer cells
for medical diagnosis.
 Factories use deep learning applications to automatically detect when people or
objects are within an unsafe distance of machines.
 You can group these various use cases of deep learning into four broad
categories— computer vision, speech recognition, natural language processing
(NLP), and recommendation engines.
Computer vision
Computer vision is the computer's ability to extract information and insights from images
and videos. Computers can use deep learning techniques to comprehend images in the
same way that humans do. Computer vision has several applications, such as the
following:
 Content moderation to automatically remove unsafe or inappropriate content
from image and video archives
 Facial recognition to identify faces and recognize attributes like open eyes,
glasses, and facial hair
 Image classification to identify brand logos, clothing, safety gear, and other
image details
Speech recognition
Deep learning models can analyze human speech despite varying speech patterns, pitch,
tone, language, and accent. Virtual assistants such as Amazon Alexa and automatic
transcription software use speech recognition to do the following tasks:
 Assist call center agents and automatically classify calls.
 Convert clinical conversations into documentation in real time.
 Accurately subtitle videos and meeting recordings for a wider content reach.

Department of CSE, GNDEC, Bidar. Page 7

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

Natural language processing

Computers use deep learning algorithms to gather insights and meaning from text data
and documents. This ability to process natural, human-created text has several use cases,
including in these functions:
 Automated virtual agents and chatbots
 Automatic summarization of documents or news articles
 Business intelligence analysis of long-form documents, such as emails and
forms
 Indexing of key phrases that indicate sentiment, such as positive and negative
comments on social media

Recommendation engines
Applications can use deep learning methods to track user activity and develop
personalized recommendations. They can analyze the behavior of various users and help
them discover new products or services. For example, many media and entertainment
companies, such as Netflix, Fox, and Peacock, use deep learning to give personalized
video recommendations.

How does deep learning work?

Deep learning algorithms are neural networks that are modeled after the human brain. For
example, a human brain contains millions of interconnected neurons that work together to
learn and process information. Similarly, deep learning neural networks, or artificial
neural networks, are made of many layers of artificial neurons that work together inside
the computer.

Artificial neurons are software modules called nodes, which use mathematical
calculations to process data. Artificial neural networks are deep learning algorithms that
use these nodes to solve complex problems.

What are the components of a deep learning network? The components of a deep
neural network are the following. Input layer
An artificial neural network has several nodes that input data into it. These nodes make
up the input layer of the system.

Department of CSE, GNDEC, Bidar. Page 8

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

Hidden layer
The input layer processes and passes the data to layers further in the neural network. These
hidden layers process information at different levels, adapting their behavior as they
receive new information. Deep learning networks have hundreds of hidden layers that
they can use to analyze a problem from several different angles.
For example, if you were given an image of an unknown animal that you had to classify,
you would compare it with animals you already know. For example, you would look at
the shape of its eyes and ears, its size, the number of legs, and its fur pattern. You would
try to identify patterns, such as the following:
 The animal has hooves, so it could be a cow or deer.
 The animal has cat eyes, so it could be some type of wild cat.

The hidden layers in deep neural networks work in the same way. If a deep learning
algorithm is trying to classify an animal image, each of its hidden layers processes a
different feature of the animal and tries to accurately categorize it.

Output layer
The output layer consists of the nodes that output the data. Deep learning models that
output "yes" or "no" answers have only two nodes in the output layer. On the other hand,
those that output a wider range of answers have more nodes.

The tampering of a digital image is called digital image forgery, these forged images
cannot be detected by the naked eye. Such images are the primary sources of spreading
fake news and misleading information in the context of society with the aid of diverse
social media platforms like Facebook, Twitter, etc. The editing software tools that can
make these forgeries are available for free with some advanced features that are used for
image tampering such as GNU, GIMP, and Adobe Photoshop. Such forgeries can be
detected using digital image forgery algorithms and techniques, these algorithms are used
in image security especially when the original content is not available. Digital image
forgery means adding unusual patterns to the original images that create a heterogeneous
variation in image properties and an unusual distribution of image features. Figure 1
shows the classification of digital image forgery. Active approaches require essential
information about the image for the verification process. The inserted information within

Department of CSE, GNDEC, Bidar. Page 9

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

the picture is employed to observe the modification in that picture. The active approach
consists of two types: digital signatures which insert some additional data obtained from
an image by the end of the acquisition process, and digital watermarking which is inserted
into images either during the acquisition phase or during the processing phase. The
passive image forgery detection methods benefit from the features retained by the image
allocation processes achieved in different stages of digital image acquisition and storage.
Passive methodologies do not require past information about the image. These approaches
exploit that the tampering actions modify the contents of information of the image that can
facilitate tampering detection. Copy move forgery involves duplicating a section or object
within an image and pasting it again in a different location within the same image to
replicate (or move) a specific scene in the image. Copy-move forgery is the most
common technique used to manipulate images, it is also the most challenging type of
forgery to detect due to the complexity of copying and replicating an object or section of
the image with identical properties and feature distributions and pasting it within the same
image. some post-processing techniques can be added after CMF processes such as
rotation, scaling, JPEG compression, etc. which makes the detection further difficult and
complex Splicing forgery can be generated by adding or blending two images or set of
images to produce an unprecedented image. The source images used to generate a spliced
image may include dissimilar color temperatures, illumination conditions, and noise levels
based on various factors. Average filtering or some other related image processing
operation can be applied as postprocessing like resizing, cropping, rotating, and
retouching each of the source images to match the visual attributes, shape, and size of the
target image so that the forged image can look realistic Retouching forgery involves
modifying an image to hide or highlight particular features such as brightness, color,
contrast, or other visual attributes and altering background coloring. It includes the visual
quality enhancement of the image. Resampling Forgery is the act of altering the
dimensionality of a particular object or section within an image to present a distorted or
misleading view. Morphing forgery involves merging two scenes from different images to
create an entirely new scene, this can be done using graphic software to create a
completely artificial image with no basis in reality. The three major types of tampering are
Copy Move, Image Splicing, and Image Retouching. Digital Image Forgery Detection is a
binary classification task, to classify the image as either forged or authentic. Recently,
deep learning has become a promising tool for enhancing digital image forgery detection.
In any Deep Learning Model, feature extraction is an important phase that affects the

Department of CSE, GNDEC, Bidar. Page 10

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

performance of the algorithm, where the database size is considered a significant factor.
Transfer learning presents a viable alternative solution when dealing with limited sample
size problems that supports taking the knowledge acquired from a previously trained
model including features, weights, and other relevant information which was trained on a
large dataset such as the ImageNet database, that contains 1.2 million images grouped
into 1000 classes to solve the problem of small size dataset in the new target domain By
utilizing a pre- trained model, significant amounts of time spent on training can be saved,
and the model can be adapted to work with smaller datasets through retraining . The
motivation behind image forgery detection is to check the authenticity of digital images,
especially when images are used as evidence in court and forensics, news, or historical
data, or in the military, and medical diagnosis systems, it prevents the distribution of
misinformation and fake news, particularly in social media and online platforms, these
forged images can be used to destroy someone‘s reputation or mislead public opinion, or
for distorting the truth in news reports, they can also be used to exaggerate the capabilities
of the countries army.

Image forgery detection has several challenges due to the nature of image manipulation
techniques. These challenges can be concluded as:

 Computational Complexity and the limitation of the CPU and memory is the main
challenge, which takes a lot of training time and most of the time runs out of memory
even with high memory specifications.
 Detecting more than one type of image forgery at the same time affects the accuracy
rate, so there is a need to improve its accuracy rate.
 There is a need to solve the problem of the accuracy-speed trade-off.
 Most image forgery detection techniques that have high detection accuracy are
very complex, there is a need for a simpler technique with high detection accuracy
rate.
 Most image forgery detection techniques suffer from detecting images that lie under
post-processing operations like image rotation, scaling, blurring, brightness
adjustment, and adding noise. This paper presents the following contribution:
 Detecting two types of passive image forgeries like image splicing and copy-move at
the same time to be suitable for real-life scenarios

Department of CSE, GNDEC, Bidar. Page 11

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

 Achieving a high accuracy rate compared to the state of art results found in the
literature. Moreover, using a pre-trained model and taking the power of transfer
learning, with a small number of parameters, the developed lightweight model is
well-suited for environments with memory and CPU limitations. This is an added
value in favor of the proposed architecture.
 Evaluating the performance of eight different pre-trained models such as VGG16,
VGG19, ResNet50, Reset101, ResNet152, MobileNetV2, Xception and DenseNet
are considered.
 A comparative analysis of the eight forementioned pretrained models and state of
art is presented.
 Using the CASIAV2 dataset which is one of the best benchmark datasets that is
considered as the main challenge itself, it contains two main types of image forgery
(splicing and copy-move) with different sizes and contains many types of images
format (TIFF, JPEG, and BMP ) and also the cropped parts in the forged images
underwent some processing including distortion, rotation, and scaling, to create an
image that seems to be real, involving blurring Image forgery detection has several
challenges due to the nature of image manipulation techniques. These challenges can
be concluded as:
 Computational Complexity and the limitation of the CPU and memory is the main
challenge, which takes a large training time and most of the time runs out of memory
even with high memory specifications.
 Detecting more than one type of image forgery at the same time affects the accuracy
rate, so there is a need to improve its accuracy rate.
 There is a need to solve the problem of the accuracy-speed trade-off.

 Most image forgery detection techniques that have high detection accuracy are
very complex, there is a need for a simpler technique with high detection accuracy
rate.
 Most image forgery detection techniques suffer from detecting images that lie
under post-processing operations like image rotation, scaling, blurring, brightness
adjustment, and adding noise. This paper presents the following contribution:
 Detecting two types of passive image forgeries like image splicing and copy-move at
the same time to be suitable for real-life scenarios.

Department of CSE, GNDEC, Bidar. Page 12

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

 Achieving a high accuracy rate compared to the state of art results found in the
literature. Moreover, using a pre-trained model and taking the power of transfer
learning, with a small number of parameters, the developed lightweight model is
well-suited for environments with memory and CPU limitations. This is an added
value in favor of the proposed architecture.
 Evaluating the performance of eight different pre-trained models such as VGG16,
VGG19, ResNet50, Reset101, ResNet152, MobileNetV2, Xception and DenseNet
are considered.
 A comparative analysis of the eight forementioned pretrained models and state of art
is presented.
 Using the CASIAV2 dataset which is one of the best benchmark datasets that is
considered as the main challenge itself, it contains two main types of image forgery
(splicing and copy-move) with different sizes and contains many types of images
format (TIFF, JPEG, and BMP ) and also the cropped parts in the forged images
underwent some processing including distortion, rotation, and scaling, to create an
image that seems to be real, involving blurring

Department of CSE, GNDEC, Bidar. Page 13

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

CHAPTER 2
LITERATURE SURVEY
2.1 Literature Survey
In the image forgery detection field, various approaches were proposed. Traditional
techniques mostly extract a set of handcrafted based features, followed by a classifying
technique like feature matching to differentiate between the authentic and forged images.
In the machine learning approach, a set of classifiers can be used in the classifying process
like Support Vector Machine and Naïve Bayes classifier. While more recent techniques
employ convolutional neural networks (CNNs) and deep neural networks (DNN)
methods, others employ the network with the help of pre-trained models and the power of
transfer learning. CNN and deep learning-based techniques will be discussed moving over
the use of different pre-trained models.

Metaheuristics with Optimal Deep Transfer Learning Based Copy- Move Forgery
Detection Technique:
ABSTRACT: The extensive availability of advanced digital image technologies and image
editing tools has simplified the way of manipulating the image content. An effective
technique for tampering the identification is copy-move forgery. Conventional image
processing techniques generally search for the patterns linked to the fake content and
restrict the usage in massive data classification. Contrastingly, deep learning (DL) models
have demonstrated significant performance over the other statistical techniques. With this
motivation, this paper presents an Optimal Deep Transfer Learning based Copy Move
Forgery Detection(ODTLCMFD) technique. The presented ODTL-CMFD technique aims
to derive a DL model for the classification of target images into the original and the
forged/tampered, and then localize the copy moved regions. To perform the feature
extraction process, the political optimizer (PO) with Mobile Networks (MobileNet) model
has been derived for generating a set of useful vectors. Finally, an enhanced bird swarm
algorithm (EBSA) with least square support vector machine (LS-SVM) model has been
employed for classifying the digital images into the original or the forged ones. The
utilization of the EBSA algorithm helps to properly modify the parameters contained in
the Multiclass Support Vector Machine (MSVM) technique and thereby enhance the
classification performance. For ensuring the enhanced performance of the ODTL-CMFD
technique, a series of simulations have been performed against the benchmark MICC-

Department of CSE, GNDEC, Bidar. Page 14

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

F220, MICCF2000, and MICC-F600 datasets. The experimental results have

demonstrated the improvised performance of the ODTL-CMFD approach over the other
techniques in terms of several evaluation measures.

A New Method to Detect Splicing Image Forgery Using Convolutional Neural

Network:
ABSTRACT: Recently, digital images have been considered the primary key for many
applications, such as forensics, medical diagnosis, and social networks. Image forgery
detection is considered one of the most complex digital image applications. More
profoundly, image splicing was investigated as one of the common types of image
forgery. As a result, we proposed a convolutional neural network (CNN) model for
detecting splicing forged images in real-time and with high accuracy, with a small number
of parameters as compared with the recently published approaches. The presented model
is a lightweight model with only four convolutional layers and four max-pooling layers,
which is suitable for most environments that have limitations in their resources. A
detailed comparison was conducted between the proposed model and the other
investigated models. The sensitivity and specificity of the proposed model over CASIA
1.0, CASIA 2.0, and CUISDE datasets are determined. The proposed model achieved an
accuracy of 99.1% in detecting forgery on the CASIA 1.0 dataset, 99.3% in detecting
forgery on the CASIA 2.0 dataset, and 100% in detecting forgery on the CUISDE dataset.
The proposed model achieved high accuracy, with a small number of parameters.
Therefore, specialists can use the proposed approach as an automated tool for real-time
forged image detection.

Image Forgery Detection Using Tamper-Guided Dual Self-Attention Network with

Multiresolution Hybrid Feature:
ABSTRACT: Image forgery detection can efficiently capture the difference between the
tampered area and the nontampered area. However, existing work usually overemphasizes
pixel-level localization, ignoring image-level detection. As a result, false detection for
tampered image maybe cause a large number of false positives. To address this problem,
we propose an end-to-end fully convolutional neural network. In this framework,
multiresolution hybrid features from RGB stream and noise stream are firstly fused to
learn visual artifacts and compression inconsistency artifacts, which can efficiently
identify the tampered images. Furthermore, a tamper-guided dual self-attention (TDSA)

Department of CSE, GNDEC, Bidar. Page 15

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

module is designed, which can focus the network‘s attention on the tampered areas and
segment them from the image by capturing the difference between the tampered area and
the nontampered area. Extensive experiments demonstrate that compared to existing
schemes, our scheme can simultaneously effectively achieve pixel-level forgery
localization and image-level forgery detection while maintaining higher detection
accuracy and stronger robustness.

Image copy-move forgery detection and localization based on super-BPD

segmentation and DCNN:
ABSTRACT: With the increasing importance of image information, image forgery
seriously threatens the security of image content. Copy-move forgery detection (CMFD)
is a greater challenge because its abnormality is smaller than other forgeries. To solve the
problem that the detection results of the most image CMFD based on convolutional neural
networks (CNN) have relatively low accuracy, an image copy-move forgery detection and
localization based on super boundary-to- pixel direction (super-BPD) segmentation and
deep CNN (DCNN) is proposed: SD- Net. Firstly, the segmentation technology is used to
enhance the connection between the same or similar image blocks, improving the
detection accuracy. Secondly, DCNN is used to extract image features, replacing
conventional hand-crafted features with automatic learning features. The feature pyramid
is used to improve the robustness to the scaling attack. Thirdly, the image BPD
information is used to optimize the edges of rough detected image and obtain final
detected image. The experiments proved that the SD-Net could detect and locate multiple,
rotated, and scaling forgery well, especially large-level scaling forgery. Compared with
other methods, the SD-Net is more accurately located and robust to various post-
processing operations: brightness change, contrast adjustments, color reduction, image
blurring, JPEG compression, and noise adding.

An efficient approach for copy-move image forgery detection using convolution

neural network:
ABSTRACT: Digital imaging has become elementary in this novel era of technology
with unconventional image forging techniques and tools. Since, we understand that digital
image forgery is possible, it cannot be even presented as a piece of evidence anywhere.
Dissecting this fact, we must dig unfathomable into the issue to help alleviate such
derelictions. Copy-move and splicing of images to create a forged one prevail in this

Department of CSE, GNDEC, Bidar. Page 16

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

monarchy of digitalization. Copy-move involves copying one part of the image and
pasting it to another part of the image while the latter involves merging of two images to
significantly change the original image and create a new forged one. In this article, a novel
slant using a convolutional neural network (CNN) has been proposed for automatic
detection of copy-move forgery detection. For the experimental work, a benchmark dataset
namely, MICC-F2000 is considered which consists of 2000 images in which 1300 are
original and 700 are forged. The experimental results depict that the proposed model
outperforms the other traditional methods for copy-move forgery detection. The results of
copy-move forgery were highly promising with an accuracy of 97.52% which is 2.52%
higher than the existing methods.

CASIA image tampering detection evaluation database.

Abstract: Image forensics has now raised the anxiety of justice as increasing cases of
abusing tampered images in newspapers and court for evidence are reported recently.
With the goal of verifying image content authenticity, passive-blind image tampering
detection is called for. More realistic open benchmark databases are also needed to assist
the techniques. Recently, we collected a natural color image database with realistic
tampering operations. The database is made publicly available for researchers to compare
and evaluate their proposed tampering detection techniques. We call this database CASIA
Image Tampering Detection Evaluation

Database. We describe the purpose, the design criterion, the organization and self-
evaluation of this database in this paper.

Deep Residual Learning for Image Recognition.

Abstract: Deeper neural networks are more difficult to train. We present a residual
learning framework to ease the training of networks that are substantially deeper than
those used previously. We explicitly reformulate the layers as learning residual functions
with reference to the layer inputs, instead of learning unreferenced functions. We provide
com prehensive empirical evidence showing that these residual networks are easier to
optimize and can gain accuracy from considerably increased depth. On the ImageNet
dataset we evaluate residual nets with a depth of up to 152 layers—8× deeper than VGG
nets [40] but still having lower complexity. An ensemble of these residual nets achieves
3.57% error on the ImageNet test set. This result won 1st place on the ILSVRC 2015
classification task.

Department of CSE, GNDEC, Bidar. Page 17

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

Deep Learning Local Descriptor for Image Splicing Detection and Localization.
Abstract: In this paper, a novel image splicing detection and localization scheme is
proposed based on the local feature descriptor which is learned by deep convolutional
neural network (CNN). A two-branch CNN, which serves as an expressive local
descriptor is presented and applied to automatically learn hierarchical representations
from the input RGB color or grayscale test images. The first layer of the proposed CNN
model is used to suppress the effects of image contents and extract the diverse and
expressive residual features, which is deliberately designed for image splicing detection
applications. In specific, the kernels of the first convolutional layer are initialized with an
optimized combination of the 30 linear high-pass filters used in calculation of residual
maps in spatial rich model (SRM) and is ne-tuned through a constrained learning strategy
to retain the high-pass filtering properties for the learned kernels. Both the contrastive
loss and cross entropy loss are utilized to jointly improve the generalization ability of the
proposed CNN model. With the block-wise dense features for a test image extracted by the
pre-trained CNN-based local descriptor, an effective feature fusion strategy, known as
block pooling, is adopted to obtain the final discriminative features for image splicing
detection with SVM. Based on the pre-trained CNN model, an image splicing localization
scheme is further developed by incorporating the fully connected conditional random eld
(CRF). Extensive experimental results on several public datasets show that the proposed
CNN based scheme outperforms some state- of-the-art methods not only in image splicing
detection and localization performance, but also in robustness against JPEG compression.

Densely connected convolutional networks.

Abstract: Recent work has shown that convolutional networks can be substantially
deeper, more accurate, and efficient to train if they contain shorter connections between
layers close to the input and those close to the output. In this paper, we embrace this
observation and introduce the Dense Convolutional Network (Dense Net), which
connects each layer to every other layer in a feed-forward fashion. Whereas traditional
convolutional networks with L layers have L connections—one between each layer and its
subsequent x0 H1 layer—our network has L(L+1) Laurens van der Maten Facebook AI
Research lvdmaaten@fb.com x1 H2 x2 H3 x3 2 direct connections. For each layer, the
feature-maps of all preceding layers are used as inputs, and its own feature-maps are used
as inputs into all subsequent layers. Dense Nets have several compelling advantages:
they alleviate the vanishing- gradient problem, strengthen feature propagation,

Department of CSE, GNDEC, Bidar. Page 18

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

encourage feature reuse, and substantially reduce the number of parameters. We evaluate
our proposed architecture on four highly competitive object recognition benchmark tasks
(CIFAR- 10, CIFAR-100, SVHN, and ImageNet). Dense Nets obtain significant
improvements over the state-of-the-art on most of them, whilst requiring less computation
to achieve high performance.

Xception: Deep Learning with Depth wise Separable Convolutions

Abstract: We present an interpretation of Inception modules in convolutional neural
networks as being an intermediate step in-between regular convolution and the depth wise
separable convolution operation (a depth wise convolution followed by a pointwise
convolution). In this light, a depth wise separable convolution can be understood as an
Inception module with a maximally large number of towers. This observation leads us to
propose a novel deep convolutional neural network architecture inspired by Inception,
where Inception modules have been replaced with depth wise separable convolutions. We
show that this architecture, dubbed Xception, slightly outperforms Inception V3 on the
ImageNet dataset (which Inception V3 was designed for), and significantly outer forms
Inception V3 on a larger image classification dataset comprising 350 million images and
17,000 classes. Since the Xception architecture has the same number of parameters as
Inception V3, the performance gains are not due to increased capacity but rather to a more
efficient use of model parameters.

A. DEEP NEURAL NETWORK BASED IMAGE FORGERY DETECTION

TECHNIQUE
DNNs can autonomously learn an extensive number of features. Over the past few years,
a variety of image forgery detection methods have been proposed, for detecting
image forgery, where many of which relied on deep learning [5]. By constructing an
appropriate neural network, deep learning networks can identify complex hidden patterns
in data and effectively distinguish the forged parts from the original image [9]. Deep
learning technique has proven to be effective in resolving many activities or issues that
machine learning algorithms were previously unable to address.

Department of CSE, GNDEC, Bidar. Page 19

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

B. PRETRAINED NETWORK-BASED IMAGE FORGERY

DETECTION TECHNIQUES
Different IFD techniques based on transfer learning will be discussed in this section. For
splicing, [18] presented multiple image-splicing forgeries using Mask R-CNN and
MobileNetV1 backbone. A novel approach utilizing ResNet50v2 was introduced in [19],
that considered image batches as an input and used YOLO CNN weights with
ResNet50v2 architecture.

2. 2 FEASIBILITY STUDY
Feasibility Study
Feasibility study examines the viability or sustainability of an idea, project, or business.
The study examines whether there are enough resources to implement it, and the concept
has the potential to generate reasonable profits. In addition, it will demonstrate the
benefits received in return for taking the risk of investing in the idea.

Types of Feasibility Study

There are several different kinds of feasibility studies. Understanding the types of
feasibility studies and the technicalities of the concept is important for any business. They
are elaborated below:

Technical Feasibility
Technical feasibility study checks for accessibility of technical resources in the
organization. In case technological resources exist, the study team will conduct
assessments to check whether the technical team can customize or update the existing
technology to suit the new method of workings for the project by properly checking the
health of the hardware and software. Many factors need to be taken into consideration
here, like staffing requirements, transportation, and technological competency.

Financial Feasibility
Financial feasibility allows an organization to determine cost-benefit analysis. It gives
details about the investment that has to go in to get the desired level of benefit (profit).
Factors such as total cost and expenses are considered to arrive simultaneously. With this
data, the companies know their present state of financial affairs and anticipate future

Department of CSE, GNDEC, Bidar. Page 20

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

monetary requirements and the sources from which the company can acquire them.
Investors can largely benefit from the economic analysis done. Assessing the return on
investment of a particular asset or acquisition can be a financial feasibility study example.

Market Feasibility
It assesses the industry type, the existing marketing characteristics and improvements to
make it better, the growth evident and needed, competitive environment of the company‘s
products and services. Preparations of sales projections can thus be a good market
feasibility study example.

Organization Feasibility
Organization feasibility focuses on the organization‘s structure, including the legal
system, management team‘s competency, etc. It checks whether the existing conditions
will suffice to implement the business idea.

Department of CSE, GNDEC, Bidar. Page 21

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

CHAPTER 3

SYSTEM REQUIREMENTS AND SPECIFICATION

3.1 SYSTEM REQUIREMENTS AND SPECIFICATION

3.1.1 Existing System and Its Disadvantages:

The existing systems for digital image forgery detection primarily utilize deep learning
techniques, with Convolutional Neural Networks (CNNs) being the most widely adopted
method for verifying image authenticity. These approaches focus on detecting specific
forensic manipulations, such as JPEG compression, contrast enhancement (CE), median
filtering (MF), additive Gaussian noise, and resampling. While CNNs have proven
effective, their high computational resource requirements and the need for extensive
datasets for training pose significant limitations. To address these challenges, transfer
learning-based deep learning methods have been introduced, leveraging pre-trained
models to reduce training time and resource demands. Existing methodologies often
incorporate preprocessing modules for noise removal and feature extraction, convolutional
modules for pattern recognition, and classification modules for forgery detection.
Benchmark datasets like BOW and BOSS Base are commonly used for evaluation.
Despite their effectiveness, existing systems lack versatility and efficiency, highlighting
the need for improved methods that balance accuracy, scalability, and computational
efficiency.

Disadvantages:

1. CNN-based methods require high computational resources and significant

training time due to their complex architecture and reliance on large datasets.

2. Existing systems focus on detecting specific forgery types, such as JPEG

compression and contrast enhancement, making them ineffective in handling
diverse real-world forgery scenarios.

Department of CSE, GNDEC, Bidar. Page 22

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

3. These methods heavily depend on specific benchmark datasets like BOW and
BOSSBase, which may not fully represent the variety of forgery techniques
encountered in practice.
4. Traditional approaches lack adaptability to multiple forgery types, restricting their
utility in dynamic and complex forgery scenarios.
5. The systems are not optimized for scalability, making it difficult to extend them
for larger datasets or more complex real-world applications.

3.1.2 Proposed System and its advantages:

The proposed system aims to develop a robust digital image forgery detection framework
capable of identifying multiple types of forgeries in real-world scenarios. Leveraging the
CASIA dataset, the system utilizes advanced deep learning architectures to detect
forgeries with high precision. The framework evaluates several state-of-the-art
convolutional neural network (CNN) models, including InceptionV3, VGG16, VGG19,
DenseNet, MobileNetV2, ResNet50, ResNet101, ResNet152, Xception, and
NASNetMobile. Each model is employed to extract features and identify subtle
inconsistencies introduced during image manipulation. Furthermore, a hybrid model is
proposed by integrating Xception and NASNetMobile, combining their complementary
strengths to enhance detection performance. The hybrid model utilizes Xception‘s
superior feature extraction capabilities and NASNetMobile‘s efficiency in processing
complex patterns, ensuring scalability and robustness. By designing a system that
supports multiple algorithms and integrating a hybrid approach, the proposed framework
addresses the limitations of existing methods and provides a comprehensive solution
for real- world digital image forgery detection challenges.
Advantages:

1. The proposed system employs multiple deep learning models, including

InceptionV3, VGG16, Xception, and NASNetMobile, ensuring the detection of
diverse forgery types.
2. By integrating Xception and NASNetMobile, the hybrid model combines superior
feature extraction and efficient pattern processing, enhancing accuracy and
robustness.
3. Leveraging the CASIA dataset provides a broader representation of forgery types,

Department of CSE, GNDEC, Bidar. Page 23

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

making the system applicable to real-world scenarios.

4. Utilizing transfer learning with pre-trained weights significantly reduces
computational demands and training time.
5. The system is designed to be scalable, allowing seamless adaptation to larger
datasets and emerging forgery detection challenges.

3.2 Functional Requirements

Data Collection: Gathering raw data from various sources like databases, APIs, or sensors
to serve as input for your machine learning model

Pre-processing: Cleaning and transforming raw data into a usable format by handling
missing values, normalizing, and encoding data to improve model performance.

Training and Testing: Feeding preprocessed data into the machine learning model to
enable it to learn patterns and relationships by adjusting its parameters and evaluating the
model's performance on unseen data to measure its accuracy and ability to generalize to
new inputs.

Modelling:Designing and selecting a machine learning algorithm or architecture to best

suit the problem and the data available.

Predicting: Using the trained model to make future predictions or classify new, unseen
data based on learned patterns.

3.3 Non-Functional Requirements

Performance: The system must achieve high accuracy and low latency in detecting digital
image forgeries, ensuring quick processing of input data and timely responses for real-
world applications.

Scalability: The architecture should be scalable to handle an increasing number of

images, diverse datasets, and emerging forgery detection techniques without degradation
in performance.

Usability: A user-friendly interface must be designed to enable seamless interaction,

ensuring accessibility for technical and non-technical users with minimal learning effort.

Department of CSE, GNDEC, Bidar. Page 24

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

Security: Robust measures must be implemented to safeguard the system against

unauthorized access, data breaches, and potential cyberattacks, ensuring the
confidentiality of sensitive data.

Reliability: The system should maintain consistent performance and accuracy under
varying operational conditions, including large-scale data processing and concurrent user
interactions.

Maintainability: The system should allow easy updates and integration of new
algorithms or features, ensuring long-term adaptability to advancements in forgery
detection techniques.

Department of CSE, GNDEC, Bidar. Page 25

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

CHAPTER 4
METHODOLOGY
The methodology for detecting digital forgery in images using transfer learning involves
leveraging pre-trained deep learning models to identify anomalies or tampered regions in
images. Initially, a large dataset of forged and authentic images is collected and
preprocessed, including resizing, normalization, and augmentation, to ensure robust
model performance. A pre-trained convolutional neural network (CNN), such as VGG,
ResNet, or Inception, is fine-tuned on this dataset. Transfer learning helps in utilizing the
features learned by these models from large-scale image datasets, significantly reducing
training time and improving accuracy. The model is adapted to detect subtle
inconsistencies in texture, lighting, or noise patterns indicative of forgery. The final layers
are modified or replaced to classify images as forged or authentic. The performance is
evaluated using metrics like accuracy, precision, recall, and F1-score. This methodology
is efficient, as it combines the power of transfer learning with domain-specific training for
high detection accuracy.

4.1 HARDWARE REQUIREMENTS:

System: intel core i5 Processor.
Hard Disk: 512 GB.
Monitor: 15-inch LED.
Input Devices: Keyboard, Mouse.
Ram: 8 GB.

4.2 SOFTWARE REQUIREMENTS:

Operating system: Windows 11. Coding Language: Python 3.10.9. Web Framework:
Flask.
Frontend: HTML, CSS, JavaScript.
Deep Learning Frameworks: TensorFlow (with Keras).

Department of CSE, GNDEC, Bidar. Page 26

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

CHAPTER 5:
SYSTEM DESIGN
5.1 SYSTEM DESIGN
The system design for detecting digital forgery images using transfer learning techniques
involves leveraging pre-trained deep learning models to enhance the detection process.
Initially, a pre-trained model, such as ResNet, VGG, or Inception, is selected for its ability
to extract high-level features from images. These models, trained on large-scale datasets
like ImageNet, provide a robust foundation by utilizing learned features that are
transferable to the image forgery detection domain. The system is then fine-tuned using a
domain-specific dataset containing various types of manipulated and authentic images.
This fine-tuning process adjusts the pre- trained model's weights to improve its sensitivity
to forgery-specific features, such as inconsistencies in texture, lighting, or edges. The
architecture also includes preprocessing techniques like resizing, normalization, and
augmentation to improve the quality and diversity of training data. Finally, the system
incorporates a classification layer to output forgery detection results, providing accurate
and efficient detection with reduced training time and improved robustness against
diverse tampering methods.

5.1.1 SYSTEM ARCHITECTURE:

The system leverages a pre-trained deep learning model, such as Res Net, VGG, or
Efficient Net, trained on a large dataset like Image Net. This pre-trained model acts as a
feature extractor, providing highly relevant and generalized features for image analysis.
The architecture is then fine-tuned with a domain-specific dataset focused on forgery
detection to adapt the model to the unique characteristics of tampered images. The input
images pass through a preprocessing stage where resizing, normalization, and
augmentation are performed to enhance robustness. These processed images are fed into
the transfer learning model, where convolutional layers extract intricate features, and
fully connected layers classify the images as forged or authentic. The system is further
optimized with techniques like dropout and regularization to prevent overfitting. Finally,
the architecture outputs forgery detection results with high accuracy, demonstrating
improved performance, reduced training time, and robustness to various tampering
methods.

Department of CSE, GNDEC, Bidar. Page 27

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

Fig.5.1.1 System architecture

DATA FLOW DIAGRAM:

A Data Flow Diagram (DFD) visually represents the flow of data within a system,
showing how information moves from inputs to outputs through processes. It helps to
identify system components, their interactions, and data transformations. In the project,
DFD is crucial for illustrating how digital images are processed, analyzed, and classified
for forgery detection. The diagram simplifies complex processes, aiding in understanding
and communication among stakeholders, ensuring clear insights into system design.
 DFD clearly shows data flow between processes, data stores, and external
entities, making system understanding easier.
 Helps visualize system interactions and data transformations, ensuring
efficient communication and design clarity among stakeholders.
 Complex systems may require multiple layers of DFD, which can lead to
increased complexity and difficulty in understanding.
 To provide a clear and structured view of how data flows through the system,
improving design and implementation.

Department of CSE, GNDEC, Bidar. Page 28

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

Import libraries

YES NO NO PROCESS
VERIFY

Exploring the dataset

Image Processing

Build & Train the model – CNN, InceptionV3,

VGG16, VGG19, DenseNet, MobileNetV2,
ResNet50, ResNet101, ResNet152, Xception,
NasNetMobile, Hybrid Model - Xception +
NASNetMobile.

User signup & Sign in

User Input

Final Outcome End process

Fig5.1.2: Data Flow Diagram

Department of CSE, GNDEC, Bidar. Page 29

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

5.2 UML DIAGRAMS

Unified Modeling Language (UML) is a standardized modeling language used to

represent the design and structure of a system through diagrams. It helps visually depict
both the static structure and dynamic behavior of a system. In the digital image forgery
detection system, UML is used to model system components, interactions, and
workflows. Diagrams such as use case, class, and activity diagrams are utilized to depict
system functionality, object interactions, and user flows, ensuring clarity in design and
enhancing communication among developers and stakeholders. UML diagrams provide a
blueprint that guides system development and troubleshooting.
Goals:

 To represent the system's architecture and design visually, making it easier for
stakeholders to understand.
 To define and structure system components, ensuring clear relationships and
interactions.
 To visualize user-system interactions, allowing effective user requirement
analysis.
 To map out system behavior, facilitating smooth flow and efficient process
execution.
 To identify potential design flaws early in the development process.
 To support team collaboration by providing a shared, standardized language.
 To ensure the system can be scaled, modified, and maintained with minimal
complexity.

Department of CSE, GNDEC, Bidar. Page 30

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

Use Case Diagram

A Use Case Diagram illustrates the interactions between users (actors) and the system,
highlighting the functionalities provided. In the context of digital image forgery detection,
it delineates how users upload images, initiate analysis, and receive results, ensuring
clarity in user-system interactions.

Fig 5.2.1: Uml Diagram

Department of CSE, GNDEC, Bidar. Page 31

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

Class Diagram
A Class Diagram represents the static structure of the system by showing its classes,
attributes, methods, and relationships. In digital image forgery detection, it models the
system's components, such as image processing modules and detection algorithms,
facilitating a clear understanding of system architecture.

Fig 5.2.2: Class Diagram

Department of CSE, GNDEC, Bidar. Page 32

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

Activity Diagram

An Activity Diagram depicts the dynamic flow of control within the system, illustrating
the sequence of activities. In digital image forgery detection, it outlines the steps from
image input, processing, analysis, to output, ensuring a clear understanding of the
detection process.

Fig 5.2.3 Activity Diagram

Department of CSE, GNDEC, Bidar. Page 33

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

Sequence Diagram

A Sequence Diagram shows how objects interact in a particular sequence, emphasizing

the order of messages exchanged. In digital image forgery detection, it details the
interactions between system components during the analysis of an image, ensuring
accurate detection processes.

Fig 5.2.4: Sequence Diagram

Department of CSE, GNDEC, Bidar. Page 34

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

Collaboration Diagram

A Collaboration Diagram illustrates the interactions between objects in a system, focusing

on the relationships and messages exchanged. In digital image forgery detection, it
highlights how different modules collaborate to process and analyze images for forgery
detection.

Fig 5.2.5: Collaboration Diagram

Department of CSE, GNDEC, Bidar. Page 35

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

Component Diagram

A Component Diagram represents the physical components in the system and their
dependencies. In digital image forgery detection, it shows how various software
components, such as image processing libraries and detection algorithms, are organized
and interact within the system.

Fig 5.2.6: Component Diagram

Deployment Diagram

A Deployment Diagram illustrates the physical deployment of artifacts on nodes, showing

the hardware and software components. In digital image forgery detection, it depicts how
the system's software components are distributed across hardware nodes, ensuring
efficient processing and analysis.

Fig 5.2.7: Deployment Diagram

Department of CSE, GNDEC, Bidar. Page 36

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

CHAPTER 6
SYSTEM TESTING

6.1 SYSTEM TESTING

The software system testing is done to check every feature and performance of the
software after all its components have been integrated. It tests if the software is working
properly as per the requirements, and it is able to solve the customers‘ needs.

The software system testing is performed to detect any issues occurring after integration
of multiple units of the software. It finds defects in the integrated software as well as in
the complete software.

The software system testing is conducted on the complete software with the perspective
of the system requirements, functional requirements, or both. It verifies the design and
characteristics of the software, and how well it satisfies the end user requirements.

Sometimes the system testing validates the software beyond the requirements mentioned
in System Requirement Specification (SRS). It is conducted by a testing team who is not a
part of the development process, and hence has an unbiased testing mindset. It is a part of
both functional and non-functional testing and is performed with the help of the black box
testing techniques.

Process of Software System Testing

The process of the the software system testing is listed below −
Step 1 − Configure the test environment where the software system testing is to be
performed.
Step 2 − Develop the software system test cases.
Step 3 − Generate the test data for running the software system test cases. Step 4 −
Execute the software system test cases and analyze the results.
Step 5 − In case of failure of the software system test cases, the defects are reported.
Step 6 − The entire regression test cases are executed to check if the existing
functionalities of the software are working as expected.
Step 7 − Report all the regression related defects. Step 8 − Retest all the fixed defects.

Department of CSE, GNDEC, Bidar. Page 37

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

Types of Software System Testing

The types of the the software system testing are listed below –

Fig 6.1.1: Types Of Software System Testing

Performance Testing

Performance Testing is performed to verify the performance, stability, reliability etc of the
software.

Load Testing
Load Testing is performed to verify the amount of load or traffic that the software can
accommodate before it crashes. It finds the threshold limit of the maximum count of users
that the software can bear at a time after its software undergoes a breakdown.
Stress Testing
Stress Testing is performed to verify if there exists any security problems leading to the

Department of CSE, GNDEC, Bidar. Page 38

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

potential scope of hacking, and other vulnerabilities. It ensures that the safety of data is
maintained while it is being exchanged between multiple units of the software. The stress
testing is done along with the Penetration Testing and user access control testing
techniques.

Scalability Testing
Scalability Testing is performed to verify the performance of the software with respect to
its capacity to scale up or down the count of the user request loads.

6.2 TEST CASES:

Table 6.2.1: Test Cases of Three Different Inputs

S.NO INPUT If available If not available

User get registered into the

1 User signup There is no process
application

2 User signin User get login into the application There is no process

Enter input for

3 Prediction result displayed There is no process
prediction

Department of CSE, GNDEC, Bidar. Page 39

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

CHAPTER 7

7.1 COMPARISION GRAPHS AND TABLES

Table 7.1.1: Performance Evaluation Table

ML Model Accuracy Precision Recall F1 -Score

0 CNN 0.972 0.976 0.989 0.982

1 lnceptionV3 0.974 0.977 0.991 0.984

2 VGG16 0.968 0.968 0.993 0.980

3 VGG19 0.974 0.977 0.990 0.984

4 DenseNet 0.942 0.935 0.996 0.964

5 MobileNetV2 0.978 0.975 0.997 0.986

6 ResNet 50 0.974 0.973 0.995 0.984

7 ResNetl Ol 0.977 0.976 0.996 0.986

8 ResNet 152 0.974 0.972 0.996 0.984

9 Xception 0.980 0.979 0.996 0.987

10 NASNetMobile 0.954 0.947 0.998 0.972

11 Extension 0.981 0.978 0.998 0.988

Department of CSE, GNDEC, Bidar. Page 40

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

COMPARISON GRAPHS:
Accuracy Comparison Graph

Figure 7.1.1 Accuracy Comparison Graph

This bar chart compares the classification accuracy of various deep learning models, with
"Extension" achieving the highest score. The accuracy scores of other models like
NASNetMobile, ResNet, and VGG are also close to optimal.

Precision Comparison Graph

Figure 7.1.2 Precision Comparison Graph

Department of CSE, GNDEC, Bidar. Page 41

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

This bar chart illustrates the precision scores of various deep learning models, with
"Extension" achieving the highest precision. Other models like NASNetMobile, ResNet,
and VGG also show comparable high performance.

Recall Comparison Graph

Figure 7.1.3 Recall Comparison Graph

This bar chart highlights the recall scores of different deep learning models, with
"Extension" attaining the highest score. Other models like NASNetMobile, ResNet, and
VGG demonstrate similarly strong recall performance.

Department of CSE, GNDEC, Bidar. Page 42

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

F1 Score Comparison Graph

Figure 7.1.4 F1 Comparison Graph

An F1 Score Comparison Graph visually represents the performance of different models

or methods by comparing their F1 scores, which balance precision and recall. It highlights
which approach achieves the best trade-off between false positives and false negatives.

7.2 SCREENSHOTS:
RUNNING OF SERVER:

Figure 7.2.1 Represents Running the server with node server.js in the Command Prompt and
access it at http://localhost127.0.0.1:5000.

Department of CSE, GNDEC, Bidar. Page 43

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

HOME SCREEN:

Figure 7.2.2 :interface or home screen after sign in or sign up

SIGN UP\SIGN IN

Figure 7.2.3 This format ensures user information is collected while maintaining clarity and simplicity.

Department of CSE, GNDEC, Bidar. Page 44

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

Figure 7.2.4 Once the user is signed up he/she can directly sign in to the page

UPLOAD IMAGE:

Figure 7.2.5: Here you can upload your image which has to be classified by clicking on choose file we can
select the image and upload it.

Department of CSE, GNDEC, Bidar. Page 45

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

SELECTING REAL IMAGE FROM FOLDER :

Figure 7.2.6.1 Once we click on the chose file we can select the picture from the created folders as shown
above in the above picture we selected the real image

Figure 7.2.6.2 Once the image is uploaded it predicts the given picture is real or fake in above figure it
detected as it is real image

Department of CSE, GNDEC, Bidar. Page 46

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

SELECTING FAKE IMAGE FROM THE FOLDER:

Figure 7.2.7.1 Once we click on the chose file we can select the picture from the created folders
as shown above in the above picture we selected the fake image

Figure 7.2.7.2 Once the image is uploaded it predicts the given picture is real or fake in above figure it
detected as it is fake image

Department of CSE, GNDEC, Bidar. Page 47

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

CHAPTER 8
CONCLUSION & FUTURE WORK
8.1 CONCLUSION
Image forgery detection has become increasingly crucial due to the widespread use of
image editing tools that enable the creation of highly convincing forged images. In this
study, we explored various deep learning models for image forgery detection, leveraging
the CASIA dataset, a widely used benchmark for evaluating image manipulation
techniques. We implemented several convolutional neural networks (CNNs), including
InceptionV3, VGG16, VGG19, DenseNet, MobileNetV2, ResNet50, ResNet101,
ResNet152, Xception, and NASNetMobile, to detect forgeries caused by various
manipulation types such as JPEG compression, contrast enhancement, and median
filtering. Each model was evaluated for its ability to accurately identify subtle
manipulations and classify the authenticity of images. Our results demonstrated that the
hybrid model, combining Xception and NASNetMobile, achieved an impressive accuracy
of 98.1%, outperforming individual models. This hybrid approach combined Xception's
superior feature extraction capabilities with NASNetMobile's computational efficiency,
providing a scalable and effective solution for real-time image forgery detection. By
integrating multiple state-of-the-art algorithms and datasets, we were able to design a
robust, high-performance system for detecting digital image forgeries, showcasing the
potential of deep learning in combating digital deception.

8.2 Future Scope:

The future scope of this work includes exploring additional deep learning models and
transfer learning techniques to further improve detection accuracy. Integrating real-time
forgery detection systems in various applications, such as social media and news platforms,
could enhance content verification. Additionally, expanding the dataset to include more
diverse types of image manipulations and testing the system in different environmental
conditions could increase robustness and generalizability.

Department of CSE, GNDEC, Bidar. Page 48

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

REFERENCES

[1] C. D. P. Kumar and S. S. Sundaram, ―Metaheuristics with optimal deep transfer

learning based copy-move forgery detection technique,‖ Intell. Autom. Soft
Comput., vol. 35, no. 1, pp. 881–899, 2023.

[2] K. M. Hosny, A. M. Mortda, N. A. Lashin, and M. M. Fouda, ―A new method to

detect splicing image forgery using convolutional neural network,‖ Appl. Sci., vol.
13, no. 3, p. 1272, Jan. 2023.

[3] F. Li, Z. Pei, W. Wei, J. Li, and C. Qin, ―Image forgery detection using tamper-
guided dual self-attention network with multi resolution hybrid feature,‖ Secur.
Commun. Netw., vol. 2022, pp. 1–13, Oct. 2022.

[4] Q.Li, C.Wang, X.Zhou, and Z.Qin, ―Image copy-move forgery detection and
localization based on super-BPD segmentation and DCNN,‖ Sci. Rep., vol. 12, no. 1,
Sep. 2022, Art. no. 14987.

[5] S. Koul, M. Kumar, S. S. Khurana, F. Mushtaq, and K. Kumar, ―An efficient

approach for copy-move image forgery detection using convolution neural network,‖
Multimedia Tools Appl., vol. 81, no. 8, pp. 11259–11277, Mar. 2022.

[6] S. Gupta, N. Mohan, and P. Kaushal, ―Passive image forensics using universal
techniques: A review,‖ Artif. Intell. Rev., vol. 55, no. 3, pp. 1629–1679, Jul. 2021.

[7] W. H. Khoh, Y. H. Pang, A. B. J. Teoh, and S. Y. Ooi, ―In-air hand gesture signature
using transfer learning and its forgery attack,‖ Appl. Soft Comput., vol. 113, Dec.
2021, Art. no. 108033.

[8] Abhishek and N. Jindal, ―Copy move and splicing forgery detection using deep
convolution neural network, and semantic segmentation‖, Multimedia Tools Appl.,
vol. 80, no. 3, pp. 3571–3599, Jan. 2021.

[9] M. M. Qureshi and M. G. Qureshi, Image Forgery Detection & Localization Using
Regularized U-Net. Singapore: Springer, 2021.

Department of CSE, GNDEC, Bidar. Page 49

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

[10] Y.Rao, J.Ni, and H.Zhao, ―Deep learning local descriptor for image splicing
detection and localization,‖ IEEE Access, vol. 8, pp. 25611–25625, 2020.

[11] K.D.Kadam, S.Ahirrao, and K.Kotecha, ‗‗Multiple image splicing dataset (MISD):
A dataset for multiple splicing,‘‘ Data, vol. 6, no. 10, p. 102, Sep. 2021.

[12] R. Agarwal, O. P. Verma, A. Saini, A. Shaw, and A. R. Patel, ‗‗The advent of deep
learning-based,‘‘ in Innovative Data Communication Technologies and Application.
Singapore: Springer, 2021.

[13] C. Haipeng, C. Chang, S. Zenan, and L. Yingda, ―Hybrid features and semantic
reinforcement network for image,‖ Multimedia Syst., vol. 28, no. 2, pp. 363–374,
2021.

[14] M. A. Elaskily, M. H. Alkinani, A. Sedik, and M. M. Dessouky, ‗‗Deep learning based

algorithm (ConvLSTM) for copy move forgery detection,‘‘ J. Intell. Fuzzy Syst.,
vol. 40, no. 3, pp. 4385–4405, Mar. 2021.

[15] A. K. Jaiswal and R. Srivastava, ―Detection of copy-move forgery in digital image

using multi-scale, multi-stage deep learning model,‖ Neural Process. Lett., vol. 54,
no. 1, pp. 75–100, Aug. 2021.

[16] A. Mohassin and K. Farida, ‗‗Digital image forgery detection approaches: A

review,‘‘ in Applications of Artificial Intelligence in Engineering. Singapore:
Springer, 2021.

[17] S. S. Ali, I. I. Ganapathi, N.-S. Vu, S. D. Ali, N. Saxena, and N. Werghi, ―Image
forgery detection using deep learning by recompressing images,‖ Electronics, vol.
11, no. 3, p. 403, Jan. 2022.

[18] K. Kadam, S. Ahirrao, K. Kotecha, and S. Sahu, ―Detection and localization of

multiple image splicing using MobileNet v1,‖ IEEE Access, vol. 9, pp. 162499–
162519, 2021.

[19] E.U.H.Qazi, T.Zia, and A.Almorjan, ―Deep learning-based digital image forgery
detection system,‖ Appl. Sci., vol. 12, no. 6, p. 2851, Mar. 2022.

Department of CSE, GNDEC, Bidar. Page 50

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

[20] A.-R. Gu, J.-H. Nam, and S.-C. Lee, ―FBI-Net: Frequency-based image forgery
localization via multitask learning with self-attention,‖ IEEE Access, vol. 10, pp.
62751–62762, 2022.

[21] K. D. Kadam, S. Ahirrao, and K. Kotecha, ‗‗Efficient approach towards detection

and identification of copy move and image splicing forgeries using mask R-CNN
with MobileNet v1,‘‘ Comput. Intell. Neurosci., vol. 2022, pp. 1–21, Jan. 2022.

[22] M. N. Abbas, M. S. Ansari, M. N. Asghar, N. Kanwal, T. O‘Neill, and B. Lee,

‗‗Lightweight deep learning model for detection of copy move image forgery with
post-processed attacks,‘‘ in Proc. IEEE 19th World Symp. Appl. Mach. Intell.
Informat. (SAMI), Jan. 2021, pp. 125–130.

[23] K. B. Meena and V. Tyagi, Image Splicing Forgery Detection Techniques: A

Review. Cham, Switzerland: Springer, 2021.

[24] N. Krishnaraj, B. Sivakumar, R. Kuppusamy, Y. Teekaraman, and A. R. Thelkar,

‗‗Design of automated deep learning-based fusion model for copy-move image
forgery detection,‘‘ Comput. Intell. Neurosci., vol. 2022, pp. 1–13, Jan. 2022.

[25] S. Jabeen, U. G. Khan, R. Iqbal, M. Mukherjee, and J. Lloret, ‗‗A deep multimodal
system for provenance filtering with universal forgery detection and localization,‘‘
Multimedia Tools Appl., vol. 80, no. 11, pp. 17025–17044, May 2021.

[26] D. Mallick, M. Shaikh, A. Gulhane, and T. Maktum, ‗‗Copy move and splicing
image forgery detection using CNN,‘‘ in Proc. ITM Web Conf., vol. 44, 2022, Art.
no. 03052.

[27] K. Simonyan and A. Zisserman, ‗‗Very deep convolutional networks for large- scale
image recognition,‘‘ 2014, arXiv:1409.1556.

[28] T.-H. Nguyen, T.-N. Nguyen, and B.-V. Ngo, ‗‗A VGG-19 model with transfer
learning and image segmentation for classification of tomato leaf disease,‘‘
AgriEngineering, vol. 4, no. 4, pp. 871–887, Oct. 2022.

Department of CSE, GNDEC, Bidar. Page 51

DETECTING DIGITAL FORGERY IMAGES USING TRANSFER LEARNING TECHNIQUE

[29] K. He, X. Zhang, S. Ren, and J. Sun, ‗‗Deep residual learning for image
recognition,‘‘ in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), Jun.
2016, pp. 770–778.

[30] G. Huang, Z. Liu, L. Van Der Maaten, and K. Q. Weinberger, ‗‗Densely connected
convolutional networks,‘‘ in Proc. IEEE Conf. Comput. Vis. Pattern Recognit.
(CVPR), Jul. 2017, pp. 2261–2269.

[31] F. Chollet, ‗‗Xception: Deep learning with depthwise separable convolutions,‘‘ in

Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), Jul. 2017, pp. 1800–
1807.

[32] R. Indraswaria, R. Rokhanab, and W. Herulambang, ‗‗Melanoma image

classification based on MobileNetV2 network,‘‘ in Proc. 6th Inf. Syst. Int. Conf.
(ISICO), 2022, pp. 198–207.

[33] J. Dong, W. Wang, and T. Tan, ‗‗CASIA image tampering detection evaluation
database,‘‘ in Proc. IEEE China Summit Int. Conf. Signal Inf. Process., Jul. 2013,
pp. 422–426.

Department of CSE, GNDEC, Bidar. Page 52