International Research Journal on Advanced Engineering e ISSN: 2584-2854

Volume: 03
and Management Issue:07 July 2025
[Link] Page No: 2425 - 2432
[Link]

Image Forgery Detection

Dr. Vikram S. Patil1, Ms. Muskan Sayyad2, Mr. Sangram Shinde3, Mr. Salman Pathan4, Mr. Shreyash Nikam5
1
Principal, Yashoda Technical Campus, Faculty of Engineering, Satara, Maharashtra, India.
2,3,4,5
UG Computer Science and Engineering, Yashoda Technical Campus, Faculty of Engineering, Satara,
Maharashtra, India.
Email ID: principalengg_ytc@[Link].in1, muskansayyad2003@gmail.com2,
sangramsinhshinde9@gmail.com3, salmanpathan.exe2003@gmail.com4, nikamshreyash23@gmail.com5

Abstract
In the digital age, the proliferation of image editing tools has made it easier than ever to manipulate images,
raising concerns about the authenticity and credibility of visual content. This project focuses on the
development of an effective and efficient image forgery detection system to address the growing challenges
associated with digital image tampering. The proposed system leverages advanced techniques in computer
vision and machine learning to detect common forms of forgeries, such as copy-move, splicing, and removal.
Using feature extraction methods such as SURF, SIFT, and deep learning models, the system identifies
inconsistencies in texture, lighting, and metadata. By employing a robust dataset of authentic and forged
images for training and testing, the system achieves high accuracy in distinguishing tampered images from
genuine ones. This project aims to contribute to areas such as digital forensics, content verification, and social
media monitoring, ensuring trustworthiness in digital media. The results demonstrate the system's potential
for real-world applications, providing an automated and reliable tool for image integrity verification.
Keywords: Image Forgery Detection, Computer Vision, Machine Learning, Copy-Move & Splicing, Feature
Extraction (SIFT & SURF), Digital Forensics.

1. Introduction
The rapid advancement of digital imaging images), and content removal (erasing or obscuring
technologies and the widespread availability of parts of an image). These manipulations often leave
sophisticated editing tools have made it easier than subtle traces, such as inconsistencies in lighting,
ever to manipulate images. While these technologies texture, or compression artifacts, which can be
offer significant benefits in creative industries, they analysed to reveal the forgery. This project aims to
also pose serious challenges to the authenticity and develop an automated image forgery detection
trustworthiness of visual content. Image forgery, system that leverages state-of-the-art techniques in
which involves the deliberate alteration of digital computer vision and machine learning. By
images, has become a widespread issue, affecting identifying and analysing telltale signs of tampering,
fields such as journalism, law enforcement, forensics, the system seeks to enhance the reliability of digital
and social media. This has created an urgent need for media and support applications in digital forensics,
reliable techniques to detect and prevent image copyright protection, and misinformation control.
tampering. Image forgery detection is the process of Through the exploration of feature- based methods
identifying whether an image has been altered and, if and deep learning models, the project contributes to
so, determining the nature and extent of the the growing efforts to combat image forgery and
manipulation. Common types of forgery include ensure the integrity of digital content. Detection of
copy-move (duplicating regions within the same such manipulations relies on identifying
image), splicing (combining parts of different inconsistencies in the digital footprint of an image,

IRJAEM 2425
 International Research Journal on Advanced Engineering e ISSN: 2584-2854
Volume: 03
and Management Issue:07 July 2025
[Link] Page No: 2425 - 2432
[Link]

such as alterations in texture, color gradients, accuracy of deep learning methods over traditional
illumination patterns, or compression artifacts. ones and presents insights into the application of
Traditional feature-based methods like SIFT (Scale- these techniques in real-world scenarios. [1] The
Invariant Feature Transform) and SURF (Speeded- paper discusses a method for detecting copy-move
Up Robust Features) have been widely used to detect image forgeries, where a part of an image is
anomalies, but they often struggle with high- duplicated and placed elsewhere in the same image.
resolution, highly manipulated images. On the other This type of forgery is difficult to detect because the
hand, deep learning approaches have shown copied parts share similar attributes with the original
significant promise, enabling the automated image. The proposed solution uses Convolutional
extraction of complex patterns and irregularities. By Neural Networks (CNNs) to extract features from
providing a systematic approach to image forgery image blocks and classify images as either original or
detection, this project aims to contribute to the forged. The model achieves an accuracy of 97.7%
growing body of tools used in digital forensics, using the CASIA2 dataset, which contains both
ensuring that visual content can be trusted in an era authentic and forged images. The approach
of pervasive digital manipulation. Furthermore, the effectively addresses challenges like image
findings of this research hold potential applications in compression and resizing, which can mask
journalism, legal systems, and social media alterations. The method offers real-time processing,
platforms, where the authenticity of images is of making it suitable for applications in social media,
paramount importance. The Image Forgery legal, and academic contexts, where image forgery
Detection project is a web-based application designed detection is critical. [2] The paper proposes a novel
to identify and analyze tampered or digitally image forgery detection system using Convolutional
manipulated images using advanced machine Neural Networks (CNNs) to identify various types of
learning and image processing techniques. In today’s image manipulations, including copy-move, splicing,
digital era, where the authenticity of visual content is and retouching. The system integrates Error Level
often questioned, this system plays a crucial role in Analysis (ELA) with deep learning to enhance
detecting forgeries by comparing image patterns, detection accuracy. Tested on real-world images, it
textures, and structural inconsistencies. The project achieved a 93% detection accuracy, outperforming
integrates technologies such as Java and Spring Boot existing methods. This CNN-based system offers a
for the backend, MySQL for secure data storage, robust solution for detecting image forgery in
Python-based ML models for analysis, and a applications like forensics, security, and digital media
responsive [Link] frontend for user interaction. By analysis, addressing the growing concern of
utilizing methods like Structural Similarity Index manipulated images in today's digital landscape.[3]
(SSIM) and Support Vector Machine (SVM) The paper Image Forgery Detection using Deep
classification, the system accurately flags suspicious Learning Model addresses the challenge of detecting
regions and presents results visually through image tampering, which can compromise the
heatmaps. integrity of visual evidence, particularly in legal and
2. Literature Survey forensic settings. The study proposes a deep learning
The paper provides a comprehensive review of deep solution using the VGG-19 architecture to identify
learning techniques for image forgery detection. It forged images. The authors utilize a dataset from
contrasts traditional methods that rely on handcrafted CASIA, containing both authentic and manipulated
features with newer deep learning approaches, which images. Preprocessing steps, including image
can automatically extract complex patterns from resizing and denoising, are applied to prepare the
images. The paper also surveys publicly available dataset for training. The model is evaluated based on
datasets for training and testing deep learning models, accuracy and loss, achieving more than 95% accuracy
comparing their effectiveness in detecting and a minimal loss value, which makes it highly
manipulated images. It highlights the improved efficient compared to other existing models. The

IRJAEM 2426
 International Research Journal on Advanced Engineering e ISSN: 2584-2854
Volume: 03
and Management Issue:07 July 2025
[Link] Page No: 2425 - 2432
[Link]

authors suggest that, in the future, the model could be deep learning approaches, that can be employed for
deployed as a tool on a website, allowing the general image forgery detection it demonstrates the
public to easily detect image forgeries. [4] The paper challenges of the current state of the work. [6] With
Image Forgery Detection Using Machine Learning the advent of powerful image editing tools,
explores the increasing concern over the authenticity manipulating images and changing their content is
of digital images due to advancements in imaging becoming a trivial task. Now, you can add, change or
technology and the widespread use of photo- editing delete significant infor mation from an image,
software. With the rise of social media and the ease without leaving any visible signs of such tampering.
of altering images, the integrity of digital images has With more than several millions pictures uploaded
been compromised, creating a need for effective daily to the net, the move towards paperless
detection methods. The study employs a machine workplaces, and the introduction of eGovernment
learning algorithm, specifically Support Vector services everywhere, it is becoming important to
Machine (SVM), to identify whether an image has develop robust detection methods to identify image
been manipulated. The model also includes a feature tampering operations and validate the credibility of
to block users who attempt to upload altered images. digital images. This led to major research efforts in
This approach helps maintain the credibility of digital image forensics for security applications with focus
images, especially in fields that rely on them for on image forgery detection and authenti cation. The
decision-making, such as medicine and warfare. [5] study of such detection techniques is the main focus
Image is a powerful way to share information in the of this paper. In particular, we provide a
digital world. The sources of images are everywhere, comprehensive survey of different forgery detection
magazines, newspapers, healthcare, entertainment, techniques, com plementing the limitations of
education, social media and electronic media. With existing reviews in the literature. The survey covers
the advancement of image editing software and cheap image copy-move forgery, splicing, forgery due to
camera-enabled mobile devices, image manipulation resampling, and the newly introduced class of
is very easy without any prior knowledge or algorithms, namely image retouching. We
expertise. So, image authenticity has questioned. particularly discuss in detail the class of pixel based
Some people use the forged image for fun, but some techniques which are the most commonly used
people may have bad intentions. The manipulated approaches, as these do not require any a priori
image may use by political parties to spread their information about the type of tampering. The paper
false propaganda. Fake images use by people to can be seen as a major attempt to provide an up-to-
spread rumours and stoking someone. In addition to date overview of the research work carried in this all
harming individuals, fake images can damage the important field of multimedia. [7] These days digital
credibility of media outlets and undermine the public image forgery has turned out to be unsophisticated
trust in them. The need for reliable and efficient because of capable PCs, propelled image editing
image forgery detection methods to combat softwares and high resolution capturing gadgets.
misinformation, propaganda, hoaxes, and other Checking the respectability of pictures and
malicious uses of manipulated images. These are identifying hints of altering without requiring
some known issues on digital images. The researcher, additional pre-embedded information of the picture
scientist, and image forensic experts are working on or pre-installed watermarks are essential examine
the development of fake image detection and field. An endeavor is prepared to review the current
identification tools. Presently digital image forgery improvements in the research area of advanced
detection is a trending field of research. The main aim picture fraud detection and comprehensive reference
of this paper is to provide the exhaustive review on index has been exhibited on passive methods for
digital image forgery detection tools and techniques. forgery identification. Passive techniques donot
It also discusses various machine learning require preembedded information in the image.
techniques, such as supervised, unsupervised, and Several image forgery detection techniques are

IRJAEM 2427
 International Research Journal on Advanced Engineering e ISSN: 2584-2854
Volume: 03
and Management Issue:07 July 2025
[Link] Page No: 2425 - 2432
[Link]

arranged first and after that their summed up initiates the detection pipeline.
organization is produced. Author will review the
various image forgery detection techniques along
with their results and also compare the various
different techniques based on their accuracy. [8] The
digital image proves critical evidence in the fields
like forensic investigation, criminal investigation,
intelligence systems, medical imaging, insurance
claims, and journalism to name a few. Images are an
authentic source of information on the internet and
social media. But, using easily available software or
editing tools such as Photoshop, Corel Paint Shop,
PhotoScape, PhotoPlus, GIMP, Pixelmator, etc.
images can be altered or utilized maliciously for
personal benefits. Various active, passive and other
new deep learning technology like GAN approaches
have made photo-realistic images difficult to
distinguish from real images. Digital image tamper
detection now focuses on determining the
authenticity and consistency of digital photos. The
major research problems use generic solutions and
strategies, such as standardized data sets,
benchmarks, evaluation criteria and generalized
approaches. This paper overviews the evaluation of Figure 1 Block Diagram
various image tamper detection methods. A brief
discussion of image datasets and a comparative study [Link] UI
of image criminological (forensic) methods are The Web User Interface (UI) serves as the front-
included in this paper. Furthermore, recently facing layer of the application. It allows users to
developed deep learning techniques along with their submit images and view results. The interface is built
limitations have also been addressed. This study aims using HTML, CSS, JavaScript, and React for a
to comprehensively analyze image forgery detection seamless and interactive experience. Upon upload,
methods using conventional and advanced deep the image is sent to the backend and simultaneously
learning approaches. [9] saved to a secure image storage database for further
3. Block Diagram processing.
The block diagram (Figure 1) illustrates the end-to- [Link] Storage
end workflow of the Image Forgery Detection This component is responsible for persistently storing
system, showcasing the interaction between the user the uploaded images. It ensures that the image data is
interface, image processing modules, machine retained throughout the detection process. It also acts
learning components, and the result visualization. as a repository for analysis logs and results, allowing
Each component in this flow plays a crucial role in retrieval for future reference or audits.
identifying and reporting image tampering. [Link] (Open CV)
[Link] Interaction Before running any detection algorithms, the image
The process begins with the user accessing the undergoes preprocessing using Open CV. This step
application through a web interface. The user uploads involves standardizing the image format, resizing,
an image they suspect may have been altered or converting to grayscale if necessary, and enhancing
forged. Once the image is uploaded, the system features such as edges and textures. These operations
prepare the image for more accurate analysis in the

IRJAEM 2428
 International Research Journal on Advanced Engineering e ISSN: 2584-2854
Volume: 03
and Management Issue:07 July 2025
[Link] Page No: 2425 - 2432
[Link]

following stages. local features, which are crucial for identifying

[Link] (SSIM) manipulated regions. For more complex forgery
The detection phase employs Structural Similarity types, deep learning models are applied to learn
Index (SSIM) to identify inconsistencies within the intricate features that are not easily captured by
image. SSIM compares different regions of the image traditional methods.
to spot anomalies in structure, brightness, and texture [Link] Detection
that may suggest tampering. This method is effective Once features are extracted, the system performs
for detecting localized changes introduced during analysis to detect tampered regions.
forgery.  Copy-Move Detection: The system uses
[Link] (SVM) block- based or keypoint-based matching to
Once potential tampered areas are identified, they are identify duplicated regions within the same
passed to a machine learning classifier. In this image. Techniques such as PatchMatch or
system, a Support Vector Machine (SVM) is used to exhaustive search are applied to detect
categorize the input as either ‘authentic’ or ‘forged’. overlapping blocks or identical feature points.
The SVM model has been trained on labeled image  Splicing Detection: Splicing often introduces
datasets to distinguish between genuine and inconsistencies in lighting and texture. The
manipulated image patterns with high accuracy. system examines discrepancies in color
[Link] (Heatmap) gradients, edge continuity, and chromatic
The final output is visualized as a heatmap, which aberrations to identify such manipulations.
overlays color-coded regions on the original image.  Metadata Analysis: The system may also
This highlights the areas most likely to be tampered. analyze image metadata (e.g., EXIF data) to
Red or warm-colored zones indicate high suspicion detect inconsistencies indicative of forgery.
of forgery, providing users with an intuitive view of [Link]
the results. The heatmap is displayed on the Web UI, The extracted features are fed into a classification
completing the user feedback loop. model, such as a Convolutional Neural Network
[Link] Report Generation (CNN) or Support Vector Machine (SVM). The
After visualization, a detailed report can be generated model is trained on a labeled dataset containing
containing the classification result, SSIM scores, and authentic and forged images, allowing it to
annotated heatmaps. This report can be downloaded distinguish between genuine and tampered regions.
or shared for documentation or forensic use. Transfer learning with pre-trained networks, such as
4. Proposed System VGG or ResNet, is used to improve detection
[Link] accuracy for complex datasets.
In this initial phase, the input image is prepared for [Link]-Processing and Visualization
analysis. Preprocessing includes converting the To improve interpretability, the system highlights
image to a standard format, resizing to a fixed tampered regions on the image, providing a visual
resolution, and enhancing features such as contrast representation of the detected forgery. The system
and brightness to improve detection accuracy. This also outputs a confidence score indicating the
step ensures consistency across the dataset and likelihood of manipulation. This feedback can be
removes potential noise that could interfere with used for further investigation or validation by
further analysis. forensic experts.
[Link] Extraction [Link] and Optimization
The system utilizes advanced feature extraction The system undergoes rigorous testing on a diverse
techniques to identify unique patterns and dataset of manipulated and genuine images. Metrics
irregularities in the image. Approaches like Scale- such as precision, recall, F1-score, and processing
Invariant Feature Transform (SIFT) and Speeded-Up time are used to evaluate performance. Based on
Robust Features (SURF) are employed to extract these metrics, the system is iteratively optimized to

IRJAEM 2429
 International Research Journal on Advanced Engineering e ISSN: 2584-2854
Volume: 03
and Management Issue:07 July 2025
[Link] Page No: 2425 - 2432
[Link]

enhance its robustness and generalizability to various [Link] Technique and Emerging
types of forgeries and datasets. Approaches
5. Working  Deepfake Detection
The image forgery detection system operates through  Multi-Modal Detection
a series of structured steps, each designed to identify  Steganalysis for Hidden Information
and analyze tampered regions in digital images. The  Forensic Watermarking
following outlines the workflow of the system.  Multi-Scale Approaches
[Link] Input and Preprocessing  Hash-Based Verification
 Input Handling: The system accepts a digital  Blind Forgery Detection
image as input in common formats such as 6. Output
JPEG, PNG, or BMP.
 Preprocessing: The image is standardized to
ensure uniformity in size and format. This
may involve resizing, grayscale conversion,
and noise reduction to enhance clarity.
Adjustments like contrast enhancement or
histogram equalization may also be applied to
improve feature visibility.
[Link] Extraction
In this step, distinctive features of the image are
extracted for analysis.
 Traditional Methods: Algorithms like SIFT Figure 2 Image Forgery Detection
(Scale-Invariant Feature Transform) and
SURF (Speeded-Up Robust Features) detect The displayed image demonstrates the output screen
keypoints and descriptors, which are critical of the “Image Forgery Detection” system after a
for identifying duplicated or altered regions. successful comparison between two images (Figure 2
 Deep Learning-Based Extraction: For more & 3). At the top, there is a user-friendly interface
complex forgery patterns, convolutional section labeled "Compare Two Images", which
neural networks (CNNs) are used to allows the user to upload two separate image files for
automatically learn spatial and texture-based analysis. In this example, the same file, [Link],
features that traditional methods might miss. appears to have been selected in both fields,
indicating either a test for self-comparison or a
[Link] Design
Choose a model depending on your forgery type and verification step. Once the user clicks the "Compare
approach. Images" button, the backend system initiates the
analysis process. A status message reading
Traditional Machine Learning:
“Comparison completed!” confirms that the image
 Extract features manually (using techniques
processing and analysis steps have finished
like SIFT, LBP, etc.).
successfully. Below this, the result section is
 Support Vector Machines (SVM)
prominently displayed. Here, a "Highlighted Image"
 Decision Trees or Random Forest is shown, featuring red rectangular boxes over
 Logistic Regression or k-NN specific areas. These red boxes indicate regions that
[Link] Learning-Based Feature Extraction the system has flagged as suspicious or potentially
Pretrained CNNs: tampered. The detection logic likely uses similarity
 Use models like VGG16, ResNet, or measurement techniques (such as SSIM) or pixel-
EfficientNet for feature extraction. level analysis to identify differences or
 Transfer learning enables leveraging large- inconsistencies in the images. Notably, some text has
scale training from other datasets. been visibly altered or obscured, and the system has

IRJAEM 2430
 International Research Journal on Advanced Engineering e ISSN: 2584-2854
Volume: 03
and Management Issue:07 July 2025
[Link] Page No: 2425 - 2432
[Link]

accurately marked these areas for user review. The friendly solution for detecting tampered content. Its
bounding boxes serve as a visual guide to help the versatility has proven invaluable across industries
user focus on questionable regions within the image. such as media, law enforcement, and research,
This interface effectively combines usability with ensuring the integrity and credibility of visual data.
technical precision, allowing users to not only As the system evolves, with enhancements in
compare visual content but also to quickly understand accuracy, performance, and ease of use, it holds
where changes might have occurred. Such a setup is immense potential for future applications like video
particularly valuable in scenarios like document forgery detection and real- time monitoring,
authentication, academic integrity checks, or digital solidifying its role as a cornerstone in digital content
forensics. verification.
Future Scope
1. Hybrid Models for Enhanced Robustness
One promising future direction is the development of
hybrid detection models that combine traditional
image processing techniques with advanced AI
methods. Traditional approaches, such as analyzing
noise patterns or frequency domains, excel at
detecting physical inconsistencies, while AI-driven
methods, especially deep learning, can adapt to
complex and diverse forgery types. By integrating
these techniques, hybrid models can overcome
Figure 3 Compare Two Images limitations of standalone methods, creating systems
that are both accurate and versatile. For instance,
The displayed image represents the output interface preprocessing an image with wavelet transforms can
of the Image Forgery Detection system, showcasing enhance subtle tampering artifacts, which deep neural
a completed analysis of an uploaded handwritten networks can then analyze for forgery detection.
document. The interface includes three key sections: 2. Real-Time Detection for Practical
the original uploaded image, the detection result Applications
image, and the final detection summary. The system As forgery detection becomes a critical need for
compares the input image against its internal forgery social media platforms, surveillance systems, and
detection algorithms using techniques like OpenCV journalism, the ability to detect tampering in real-
preprocessing and SSIM analysis. In this instance, the time is essential. However, the computational
result indicates the image is "Real" with 0.00% demands of many current algorithms make this
forged area, meaning no signs of tampering were challenging. The future lies in developing
found. The highlighted image confirms this outcome, lightweight, efficient algorithms capable of operating
as it shows no red-marked or suspicious regions. This on edge devices like smartphones and cameras. Such
result panel provides users with a clear and systems would allow instant forgery detection during
trustworthy verification of image authenticity, image uploads or live surveillance, ensuring content
combining visual evidence and statistical accuracy in authenticity without delays. This real-time capability
a user-friendly layout. could also be a game-changer in live broadcasting,
Conclusion preventing the dissemination of tampered visuals.
The Image Forgery Detection System emerges as an 3. Integration with Augmented Reality (AR)
essential tool in combating the rising challenge of and Virtual Reality (VR)
digital image manipulation in today's media-driven As AR and VR technologies become mainstream,
world. By integrating advanced machine learning detecting manipulations in immersive environments
algorithms with forensic analysis techniques, the will be crucial. Future forgery detection systems
system provides a reliable, efficient, and user-

IRJAEM 2431
 International Research Journal on Advanced Engineering e ISSN: 2584-2854
Volume: 03
and Management Issue:07 July 2025
[Link] Page No: 2425 - 2432
[Link]

could analyze 3D content for anomalies, such as review of digital forensics approaches, April
inconsistent textures, lighting, or geometry. 2024. Authors: Satyendra Singh & Rajesh
Integrating forgery detection with AR/VR tools could Kumar.
enable users to overlay detection results on [7]. A bibliography of pixel-based blind image
immersive content, helping identify tampered forgery detection techniques. August 2015,
elements in real-time within virtual spaces. Muhammad Ali Qureshin, Mohamed Deriche
4. Explainable AI for Forensics [8]. A review paper on digital image forgery
One limitation of many current AI-based detection detection techniques July 2017. Authors:
systems is their black-box nature, which can Navpreet Gill, Amit Deogar.
undermine trust and reliability in critical scenarios [9]. Comprehensive analyses of image forgery
like legal investigations. A future focus on detection methods from traditional to deep
explainable AI (XAI) can address this by making learning approaches: an evaluation, October
detection models interpretable. Systems that not only 2022. Authors: Preeti Sharma, Manoj Kumar,
identify forgery but also provide visual and textual Hitesh Sharma.
explanations for their findings—such as highlighting
tampered regions or inconsistent metadata—will be
more transparent and credible. This feature would
enhance their applicability in courtrooms and other
formal settings.
5. Proactive Detection with Embedded AI in
Cameras
Embedding forgery detection algorithms directly into
camera hardware is a proactive future direction. Such
systems could analyze an image at the moment of
capture, detecting any unauthorized alterations or
inconsistencies. For example, AI-enabled cameras
could flag potential tampering when saving an image
or video.
Reference
[1]. Image Forgery Detection using Deep
Learning: A Survey, April 2020 Authors: -
Zankhana J. Barad, Mukesh M. Goswami
[2]. Image Forgery Detection, June [Link]:
Dipanshu Narayan, Himanshu, Rishabh
Kamal.
[3]. Image Forgery Detection using CNN, August
2023. Authors: Meet Patel, Kartikay Rane,
Niyati Jain, Praneel Mhatre, Shree Jaswal .
[4]. Image Forgery detection using Deep Learning
Model, November 2022. Authors: Praveen
Gupta, Chour Singh Rajpoot, [Link],
Dvsssv Prasad, AshokKumar, S Sandeep
Kumar.
[5]. Image Forgery detection using Deep Learning
Model, December 2022.
[6]. Image forgery detection: comprehensive

IRJAEM 2432