Adama Science & Technology University
School of Electrical Engineering and Computing
MSc in Computer Science and Engineering (AI)
Second year
Seminar in AI. (CSEg 7201)
Application of Explainable AI in Detecting Forgery
Educational Certificate using Deep Learning
Nesrudin Abdulwehab Aliye
Submitted to: - Teklu Urgessa
(PhD, Associate Professor)
January, 2024
Acknowledgement
First and foremost, I would like to thank our almighty God for giving me the strength,
knowledge, ability and opportunity to undertake this research study and to persevere
and complete it satisfactorily. Without his blessings, this achievement would not have
been possible.
Next, my deepest gratitude goes to my advisor Teklu Urgessa (PHD) for his
continuous followup, guidance, suggestions and his constructive comments were very
essential in the successful completion of my duty.
I would like also to thank my family specially my dad Haj-Abdulwehab Aliye Jemal
for supporting me in all of my study and for giving me financial and moral support to
achieve my work.
Last but not least, I would like to thank our collegues/classmate friends Wondossen
Argaw and Wondimu Getachew for their contribution directly and indirectly for
success of my study.
I
Table of contents
Acknowledgement .......................................................................................................................I
Table of contents ........................................................................................................................ II
Proposal Summary .................................................................................................................... III
CHAPTER ONE .................................................................................................................... - 1 1. Introduction ........................................................................................................................ - 1 1.1. Background of the study ................................................................................................. - 1 1.2. Motivation of the study ................................................................................................... - 3 1.3. Statement of the problem ................................................................................................ - 3 1.4. Research Questions ......................................................................................................... - 4 1.5. Objectives of the study: ...................................................................................................- 4 1.5.1. General Objectives: ......................................................................................................- 4 1.5.2. Specific Objectives: ..................................................................................................... - 4 1.6. Significance and Expected outcome of the study ........................................................... - 5 1.6.1. Significance of the study ..............................................................................................- 5 1.6.2. Expected outcome, Scope and Limitation of the study ................................................- 5 CHAPTER TWO ................................................................................................................... - 7 2. Literature Review ...............................................................................................................- 7 2.1. Related works ..................................................................................................................- 7 2.2. Literature Review on Detecting Forgery Educational Certificate ...................................- 9 2.2.1. Summary of reviewing related works ........................................................................ - 13 2.2.2. Gap Analysis .............................................................................................................. - 16 CHAPTER THREE ..............................................................................................................- 18 3. Methodology .................................................................................................................... - 18 3.1. Research Methods ......................................................................................................... - 18 3.2. Research Design ............................................................................................................- 19 3.3. Materials ........................................................................................................................- 19 3.4. Procedures ..................................................................................................................... - 21 3.5. Evaluation ..................................................................................................................... - 22 3.6. Limitations .................................................................................................................... - 23 3.7. Conceptual framework .................................................................................................. - 24 CHAPTER FOUR ................................................................................................................- 27 4. Work and Budget Plan ..................................................................................................... - 27 4.1. Work plan ......................................................................................................................- 27 4.2. Budget plan ................................................................................................................... - 28 4.3. Future Work .................................................................................................................. - 29 References: ...........................................................................................................................- 31 -
II
Proposal Summary
Forgery of educational certificates is a serious issue that can have far-reaching
consequences. To detect and prevent forgery, educational institutes issuing academic
certificates or transcripts must meet certain requirements to ensure the authenticity of
the certificates. One method of detecting forgery and fabrication in educational
certificates and academic transcripts is by using cryptography and QR codes. This
involves converting certificate details into a ciphertext, encoding it in a QR code, and
printing the same on the certificate or transcript to be issued. The certificate owners,
other institutes, employers, and third-party verifiers can verify certificates for forgery
and fabrication using the scanning app of the issuing institute. Another method to
prevent certificate fraud is by issuing digital certificates with blockchain
technology. This ensures that certificates cannot be forged and verifiers can instantly
know if they are fake or not.
Explainable artificial intelligence (XAI) is a set of processes and methods that allows
human users to comprehend and trust the results and output created by machine
learning algorithms. It helps characterize model accuracy, fairness, transparency and
outcomes in AI-powered decision making. Explainable AI is crucial for an
organization in building trust and confidence when putting AI models into production.
Bias, often based on race, gender, age or location, has been a long-standing risk in
training AI models. Further, AI model performance can drift or degrade because
production data differs from training data. This makes it crucial for a business to
continuously monitor and manage models to promote AI explainability while
measuring the business impact of using such algorithms. Explainable AI can help
humans understand and explain machine learning (ML) algorithms, deep learning
and neural networks. It also promotes end user trust, model auditability and
productive use of AI. It mitigates compliance, legal, security and reputational risks of
production AI.
As AI becomes more advanced, humans are challenged to comprehend and retrace
how the algorithm came to a result. There are many related works on explainable
implementation of artificial intelligence in detecting forgery educational certificates.
A literature review is a comprehensive summary and analysis of the existing research
on a topic. The proposed system contain two methods to detect the fake documents.
First, the QR-code scanner which scan the QR-code of the document and detect that
document is original or fake. Second, the image processing techniques undergoes
three stages: training phase, testing phase, classification to detect the fake documents.
In this proposed project, the originality of document is discussed and focused on
making the detection of forgery document more robust and reliable. By the Neural
network and error value analysis algorithm using image processing system to detect
the forgery document.
Key Words: - Explainable AI, Educational Certificate, machine learning algorithms, deep
learning, neural networks, Forgery Detection, Cryptography, QR Code.
III
CHAPTER ONE
1. Introduction
1.1. Background of the study
Explainable AI (XAI) can play a crucial role in detecting forged educational certificates by
enhancing transparency, accuracy, and user trust in the verification process. The text
“Applications of Explainable AI in Detecting Forgery in Educational Certificates” refers to
the use of artificial intelligence (AI) that is understandable and interpretable by humans
(Explainable AI) in identifying fraudulent educational certificates.
This involves using AI algorithms to analyze educational certificates and detect any
anomalies or discrepancies that might indicate forgery. The “explainable” aspect means that
the AI provides clear, understandable reasons for its decisions, making it easier for humans to
trust and verify the AI’s findings. For instance, one method involves using cryptography and
QR codes. The details of the certificate are converted into a ciphertext, encoded in a QR code,
and printed on the certificate. Certificate owners, other institutes, employers, and third-party
verifiers can then verify the certificates for forgery using the scanning app of the issuing
institute.
Explainable AI (XAI) is a set of processes and methods that allows human users to
comprehend and trust the results and output created by machine learning algorithms. XAI is
used to describe an AI model, its expected impact and potential biases. It helps characterize
model accuracy, fairness, transparency and outcomes in AI-powered decision making.
Another approach uses image processing algorithms and feature point matching tools to
identify suspected irregular patterns in the certificate. Yet another method uses the Gray
Level Co-occurrence Matrix (GLCM) and Singular Value Decomposition (SVD) algorithm to
detect forged scans of educational certificates.
These applications of explainable AI not only help in maintaining the integrity of the
educational system by preventing the use of forged certificates but also ensure transparency
in the process. XAI can be used to improve the fairness and transparency of educational
assessments. For instance, XAI can help explain how an AI system arrived at a particular
grade or score, making it easier for students to understand and trust the system’s findings.
Using fake certificates can have serious consequences. If an employer finds out that an
employee has submitted a fake certificate, they can take legal action against the
-1-
employee. The potential consequences could include termination of employment, legal fines,
and even imprisonment in some cases.
Fake credentials have become a global problem. Falsified academic credentials are a serious,
prevalent, and ever-increasing problem. In 2015, the New York Times reported on a billiondollar industry consisting of 3,300 “diploma mills” that sold fake certificates for all levels of
degrees worldwide. Buying totally fake academic certificates is only part of the problem.
Those who have degrees may falsify their academic transcripts. This is made easier by the
availability of sophisticated technology. Education, as Nelson Mandela best described it, is
the most potent weapon that can change the world. Thus, the need to provide quality
education with appropriate inspection and control systems is crucial. And one of the
mechanisms of doing this is tackling the prevalence of fake documents.
The government of Ethiopia has spotted more than 200,000 fake degree certificates, the
majority of which are offered to government officials. According higher education quality
assurance agency it has traced a university college which had been offering up to 15,000 fake
master degrees in a year while other had given counterfeit degrees even after it was
closed. According to a study conducted by the Government Expenditure Management and
Control Office of Ethiopia some regional offices will badly be affected if the government
starts invalidation. The higher education quality assurance agency blamed government
officials and justice institutions for collaborating with the university colleges which have
been issuing fake certificates. The higher education quality assurance agency said it would
taking “decisive action to crack down on degree fraud” that “cheats genuine learners”.
Academic credential forgery, also known as diploma mill fraud, is the creation or use of false
educational documents to gain employment, higher education, or other benefits. This can
include counterfeiting diplomas, transcripts, and other official documents from accredited
institutions.
Fake qualifications pose a reputational risk to universities and employers. It undermines their
legitimacy and reputation and robs honest candidates of opportunities for further education or
employment. Fortunately, there are steps that universities and employers can take to protect
themselves. These include the use of verification systems, reference checking, and
competency-based interviews.
-2-
1.2. Motivation of the study
Having an occupational certification, which proves your compliance with minimal
competency criteria, can be a crucial prerequisite for securing employment in regulated
industries.
These days, Ethiopia is home to a large number of young individuals with diverse skill sets.
Since a large number of professionals have requested occupational assessments over the
previous sixteen years. Massive volumes of professional data are entered into the database
annually. But as an information technology specialist, I noticed that a number of candidates
had failed in many domains when looked at the candidate data. For this reason, it's interesting
to analyze and assess the problem domain. That claims that because there is sufficient data
for better analysis and decision-making, proper attention must be paid to the assessment and
certification processes.
1.3. Statement of the problem
Educational institutions increasingly rely on digital certificates to verify credentials. However,
forgeries of these certificates are a growing concern, posing a threat to academic integrity and
employment opportunities. Deep learning models have shown promising results in detecting
forged certificates, but their internal workings are often opaque, making it difficult to
understand why a particular certificate is flagged as fraudulent.
Existing deep learning approaches for certificate forgery detection lack explainability,
making it difficult to understand why a model classifies a certificate as genuine or forged.
This opacity raises concerns about fairness, accountability and interpretability. This research
aims to bridge this gap by applying Explainable AI (XAI) techniques to deep learning models
for educational certificate forgery detection. We will develop a system that not only achieves
high accuracy but also provides clear explanations for its classifications.
To address these challenges and ensure the integrity of workforce certification system, a more
efficient, reliable, and transparent approach to detecting forged educational certificates is
necessary. This is where Explainable Artificial Intelligence (XAI) and deep learning offer a
promising solution.
-3-
1.4. Research Questions
The questions on the Application of Explainable AI in Forgery Certificate in artificial
intelligence (AI) include:
1. How can machine learning algorithms be effectively trained to identify various types
of forgery in educational certificates (e.g., altered text, fake stamps, paper
inconsistencies)?
2. What features and patterns within certificates are most effective for distinguishing
forgeries from genuine documents?
3. How can we ensure the robustness of such systems against potential adversarial
attacks from sophisticated forgers?
1.5. Objectives of the study:
1.5.1. General Objectives:
This research aims to investigate the application of Explainable Artificial Intelligence (XAI)
techniques for detecting forgeries in educational certificates. The focus will be on
understanding how XAI can improve the interpretability and trustworthiness of automated
forgery detection systems.
1.5.2. Specific Objectives:
The specific objectives of detecting forgery educational certificates are to:
Develop accurate and scalable machine learning models that can effectively identify
various types of forgery in certificates (e.g., altered text, fake stamps, paper
inconsistencies).
Identify key features and patterns within certificates that serve as reliable indicators of
forgery.
Ensure the robustness of detection systems against potential adversarial attacks from
skilled forgers.
Explore and integrate other relevant technologies like image analysis, forensics
techniques, and blockchain technology to enhance detection accuracy and security.
-4-
1.6. Significance and Expected outcome of the study
1.6.1. Significance of the study
Detecting forgery in educational certificates is a crucial issue that has far-reaching
consequences. Fake academic certificates are becoming increasingly common, and they can
undermine the integrity of the education system. In order to protect society from these
fraudulent documents, digital systems can play a crucial role. One approach to detecting
forged scan certificates is to use image processing techniques such as GLCM and SVD
algorithms. Another approach is to use cryptography and QR codes to prevent forgery and
fabrication in educational certificates and academic transcripts.
The significance of detecting forgery in educational certificates is multifold. It helps to:
Ensure the authenticity of academic credentials.
Protect the integrity of the education system.
Prevent fraud and misrepresentation in job applications.
Ensure that only qualified candidates are hired for jobs.
Protect the reputation of educational institutions.
1.6.2. Expected outcome, Scope and Limitation of the study
The expected outcome of detecting forgery in educational certificates is to ensure the
authenticity of academic credentials, protect the integrity of the education system, prevent
fraud and misrepresentation in job applications, ensure that only qualified candidates are
hired for jobs, and protect the reputation of educational institutions.
Scope:
This study will focus on developing and evaluating a system for detecting forged educational
certificates using a combination of deep learning and explainable AI (XAI) techniques.
Deep Learning Model Development:
o
The study will explore and train deep learning models, likely convolutional neural
networks (CNNs) or transformers, to classify educational certificates as genuine or
forged.
o
Data for training and testing the model will be collected from a defined source (e.g.,
educational institutions, public datasets). The scope will encompass various types of
educational certificates (e.g., diplomas, degrees) but may be limited to a specific
region or issuing body depending on data availability.
-5-
Explainable AI Integration:
The study will integrate XAI techniques like LIME or SHAP to make the deep
o
learning model's decision-making process interpretable.
The focus will be on explaining the model's predictions for individual certificate
o
images, highlighting the regions of interest (ROIs) that influence its classification
(genuine/forged).
Evaluation and Analysis:
The model's performance will be evaluated using standard metrics like accuracy,
o
precision, recall, and F1-score on a separate testing dataset.
The effectiveness of the XAI techniques in providing human-interpretable
o
explanations for the model's decisions will be assessed.
Limitations:
Data Availability: The quality and quantity of training data significantly impact the
model's performance. The study's scope might be limited by the availability of a diverse
and comprehensive dataset of genuine and forged educational certificates.
Forgery Techniques: The model's ability to detect forgeries will depend on the types of
forgeries included in the training data. The scope might not encompass all possible
forgery techniques, such as sophisticated manipulations undetectable by the chosen deep
learning architecture.
Explainability Granularity: XAI techniques offer insights into the model's decisions,
but the level of detail and interpretability might be limited. The study might not achieve
perfect human-like explanations for every prediction.
Generalizability: The model's performance might vary depending on the format and
layout of certificates from different institutions. The scope might be limited to a specific
type or region of educational certificates, requiring further adaptation for broader
application.
Additional Considerations:
This study focuses on the technical aspects of forgery detection. Ethical
considerations regarding data privacy, potential biases in the training data, and the
responsible use of such a system will need further exploration.
The study lays the groundwork for a practical system. Integrating the model with
existing verification processes and user interface development would fall outside the
immediate scope but could be addressed in future work.
-6-
CHAPTER TWO
2. Literature Review
2.1. Related works
Educational certificates are vital documents used for academic verification and career
opportunities. Unfortunately, there's an increasing problem with forged certificates, where
individuals tamper with the content to misrepresent their qualifications. Detecting Forgery in
Educational Certificates using Deep Learning involves utilizing the power of artificial neural
networks, specifically deep learning models, to automatically identify fake or altered
educational certificates.
The application of deep learning for document forgery detection has gained significant
traction in recent years. However, incorporating explainable AI (XAI) techniques into such
systems is a relatively new area of exploration.
Deep Learning for Document Forgery Detection:
V. Christlein et al. (2018): This work explores Convolutional Neural Networks (CNNs) for
handwritten signature forgery detection. They achieve high accuracy but acknowledge the
"black-box" nature of deep learning models and call for explainability techniques.
Y. Liu et al. (2020): This paper investigates CNNs for diploma forgery detection in China.
They achieve promising results but emphasize the need for interpretability to gain trust from
users.
S. Roy et al. (2021): Here, the authors propose a framework using transfer learning with pretrained CNNs for various document forgery detection tasks. Their work highlights the
importance of explainability for real-world applications.
Explainable AI (XAI) for Document Analysis Tasks:
M. Tessler et al. (2020): This work investigates LIME (Local Interpretable Model-Agnostic
Explanations) for explaining decisions made by deep learning models in document image
classification tasks. Their findings demonstrate the potential of LIME for understanding these
models.
-7-
A. Romero et al. (2018): The authors explore SHAP (SHapley Additive exPlanations) for
explaining predictions made by deep models on text classification tasks. Their work suggests
the applicability of SHAP for various document analysis problems.
J. Xu et al. (2021): This paper proposes a combined approach using LIME and Grad-CAM
(Gradient-weighted Class Activation Mapping) for explaining forgeries in historical
handwritten documents. Their work showcases the benefits of combining multiple XAI
techniques.
Explainable AI in Educational Credential Verification:
While research directly combining explainable AI, deep learning, and educational certificate
forgery detection is limited, there's growing interest in explainability for AI-powered
credential verification systems.
For instance, the work by A. Greenhalgh et al. (2021) explores explainability for AI-based
diploma verification, highlighting the importance of transparency and trust in such systems.
The application of Explainable AI (XAI) and deep learning for detecting forged documents,
including educational certificates, is a growing area of research.
Deep Learning for Document Forgery Detection:
Xu et al. (2020): This study proposes a Convolutional Neural Network (CNN) based
approach for detecting forgeries in ID cards. Their model achieved high accuracy in
identifying manipulated regions of the document.
Feng et al. (2019): This research explores the use of Recurrent Neural Networks (RNNs) for
analyzing text inconsistencies and layout irregularities in forged certificates. Their approach
demonstrated promising results for textual forgery detection.
Hou et al. (2018): This work investigates the application of deep learning for forgery
detection in various document types, including certificates. They highlight the importance of
using large and diverse datasets to train effective models.
Explainable AI (XAI) for Document Analysis:
Singh et al. (2022): This research explores techniques for explaining the decision-making
process of deep learning models used for document classification tasks. Their work highlights
the importance of XAI for building trust and understanding in such systems.
-8-
Zhang et al. (2021): This study proposes an XAI approach specifically designed for image
classification tasks, which can be applied to interpreting forgery detection models that rely
on image analysis of certificates.
Liu et al. (2020): This work explores various XAI methods for explaining deep learning
models used in anomaly detection tasks. Detecting forged certificates can be framed as an
anomaly detection problem, and these techniques can be adapted for this specific use case.
Limitations and Considerations:
Generalizability of Deep Learning Models: Deep learning models can be sensitive to
data variations. It's crucial to train the model on a dataset that reflects the specific
characteristics of educational certificates issued in Oromia to ensure generalizability and
avoid biases.
Interpretability vs. Accuracy: There might be a trade-off between achieving high
accuracy and obtaining perfectly interpretable explanations from the deep learning
model. XAI techniques can help find a balance between these two aspects.
Integration with Workflow: Consider how the Explainable AI and deep learning system
will be integrated into existing verification processes for optimal workflow efficiency
and user adoption.
By understanding the existing research and addressing potential limitations, your project can
leverage the strengths of Explainable AI and deep learning to develop a robust and userfriendly system for detecting forged educational certificates within the system.
2.2. Literature Review on Detecting Forgery Educational Certificate
Explainable Artificial Intelligence (XAI) has gained significant attention due to the
increasing use of AI systems across various industries, including education. In educational
settings, building trust and ensuring the efficacy of AI systems require transparency and
interpretability. Detecting forgery in educational certificates is a critical task, and XAI plays a
crucial role in achieving this. Explainable AI holds promise in detecting forgery in
educational certificates. By enhancing transparency and interpretability, we can ensure the
integrity of educational credentials.
Challenges of XAI in Education
1.
Complexity of AI Algorithms: AI algorithms can be intricate and difficult to understand.
XAI aims to make these algorithms more transparent and interpretable.
-9-
2.
Necessity for Transparency: In educational contexts, decisions made by AI systems can
significantly impact students’ learning outcomes and prospects. Transparency is essential
to build trust and accountability.
Solutions and Approaches
1.
Human-AI Collaboration: Researchers have explored collaborative approaches where
AI systems work alongside humans. This ensures that decisions are explainable and can
be validated.
2.
Explainability Techniques: Various methods, such as feature importance analysis,
attention mechanisms, and rule-based models, enhance the interpretability of AI systems.
3.
Ethical and Legal Frameworks: Establishing guidelines and regulations for AI use in
education ensures responsible and transparent practices.
Competencies and Skills
1.
Student and Educator Training: Developing competencies and skills to interact
effectively with AI is crucial. Educators and students should understand how AI systems
function and interpret their decisions.
2.
Impact on Politics and Government: XAI has implications beyond education.
Policymakers need to consider the ethical and legal aspects of AI adoption.
Educational certificates are crucial documents that verify academic achievements and
qualifications. However, the growing importance of these certificates has unfortunately led to
an increase in forgery attempts. This literature review explores various methods employed to
detect forged educational certificates.
Traditional Techniques:
Visual Inspection: This initial method relies on trained personnel to identify
inconsistencies in formatting, fonts, seals, watermarks, and printing quality. While
effective for some forgeries, it can be time-consuming and susceptible to human error.
Ultraviolet (UV) Light Detection: Many genuine certificates contain security
features invisible to the naked eye but detectable under UV light. However, forgers
can replicate these features, limiting their effectiveness.
- 10 -
Automated Techniques:
The rise of digital documents and readily available editing tools necessitates automated
forgery detection methods.
Image Processing Techniques:
o
Color Analysis: Statistical analysis of color distribution can identify
inconsistencies suggestive of forgery.
o
Font Analysis: Features like font type, size, and kerning (spacing between
characters) can be compared against a database of legitimate fonts used by
specific institutions.
o
Texture Analysis: Techniques can detect inconsistencies in the texture of the
paper or printing, potentially revealing modifications.
Machine Learning (ML) and Deep Learning (DL): These techniques offer
powerful tools for automated forgery detection.
o
Supervised Learning: Models are trained on datasets of genuine and forged
certificates. During testing, the model analyzes unseen certificates and
classifies them as genuine or forged.
o
Deep Learning: More complex neural networks can learn intricate features
from image data, potentially achieving higher accuracy than traditional ML
methods.
Challenges and Considerations:
Evolving Forgery Techniques: As detection methods improve, forgers develop new
techniques to bypass them. Continuous adaptation of detection algorithms is essential.
Data Availability: Training accurate ML and DL models requires a large and diverse
dataset of both genuine and forged certificates. Acquiring such data can be
challenging due to privacy concerns and the difficulty in obtaining verified forged
examples.
Explainability of AI Models: While powerful, black-box AI models can be difficult
to interpret. Understanding why a model classifies a certificate as forged is crucial for
building trust in the system.
- 11 -
Detecting forged educational certificates is a pressing concern in today's globalized world,
impacting individuals, institutions, and society as a whole. This review aims to provide an
overview of existing research in this field, highlighting key approaches, findings, and
challenges.
A Machine learning approaches show promising results, with some studies reporting
accuracy exceeding 95% is one of the key findings. Combining multiple modalities, such as
text analysis, image analysis, and metadata checks, can further improve detection
performance. Open-source datasets and collaborative research efforts are also crucial for
accelerating progress and addressing evolving forgery techniques.
Challenges and Future Directions are bias and fairness, ethical considerations and ability to
new forgery techniques. Training data and algorithms need careful curation to avoid biases
against specific institutions or groups. Balancing data privacy with effective verification
requires transparent and secure data handling practices. Detection systems need to be
continuously updated to remain effective against evolving threats.
Detecting forged educational certificates requires a multi-pronged approach. While traditional
methods still play a role, emerging technologies like machine learning and blockchain offer
promising solutions for achieving greater accuracy, efficiency, and transparency. Addressing
the challenges related to bias, ethics, and adaptability is crucial for ensuring the long-term
effectiveness of these solutions. Continued research and collaboration among researchers,
institutions, and policymakers are essential in protecting the integrity of academic credentials
and upholding the value of genuine qualifications.
- 12 -
2.2.1. Summary of reviewing related works
Author(s) and Title
Methodology
(Algorithms)
Experiment Result
Research Questions
Research Gap
Mrs. G. Chandra Praba,
E. Jeevitha, A. Abitha, A. Abitha, A.
Abitha. “Fake Education Document
Detection using Image Processing and
Deep Learning”
- QR-code scanner
- image processing
techniques
Combination of Image
Processing works together
very efficiently and the
results obtained are accurate
Less effective in
countering the danger of faking
identity documents in the current
technology
Ali, Syed Sadaf, Iyyakutti Iyappan
Ganapathi, Ngoc-Son Vu, Syed
Danish Ali, Neetesh Saxena, and
Naoufel Werghi. “Image Forgery
Detection Using Deep Learning by
Recompressing Images”
Convolutional neural
networks
(CNNs)
Isizoh A.N., Anyi D.O., Onyeyili
T.C., Ebih U.J., Ejimofor I.A.
“Certificate Fraud Detection Using
Artificial Intelligence Technique”
Self defining
equations and
modeling diagrams
- The proposed technique can
efficiently detect image
splicing and copy-move
types of image forgeries.
- The experiment results are
encouraging, with an overall
validation accuracy of
92.23%.
The result recorded achieved
a Mean Square Error (MSE)
performance of 0.000100Mu
and Regression value of R=
0.99373 which is very good,
with implication that the new
system is very reliable.
What are the most effective
methods to detect forgery in
official documents?
How do these methods
compare to each other in terms
of accuracy and efficiency?
What are the most effective
methods to detect unseen
forgeries in an image?
How does the proposed system
compare with other methods in
terms of detecting unseen
forgeries in an image?
The lack of generalizability of
existing certificate verification
systems.
The limitations of conventional
systems and the need for automated
solutions.
Kethepalli Mallikarjuna, Pushan
Kumar Dutta, N. Sateesh Kedarnath
Kumar. “Detecting Forged Scan of
Educational Certificates Using a New
Feature Set Matching Algorithm”
Image processing
algorithm
How can deep learning-based
systems be improved to detect
a wider range of image
forgeries?
How does the proposed system
compare with other methods in
terms of detecting unseen
forgeries in an image?
How can we enhance
trustworthiness and
authenticate images by
detecting forged scan
certificates using feature point
matching algorithm and
adaptive segmentation?
- 13 -
- Image forgery localization
- Lack of performance in terms of
accuracy and reduce their time
complexity
The research paper does not
mention any research gap. It is
possible that the authors did not
consider the need for a research gap
in their study.
Author(s) and Title
Methodology
(Algorithms)
Experiment Result
Research Questions
Research Gap
Zanardelli, M., Guerrini, F., Leonardi,
R. et al. Image forgery detection: a
survey of recent deep-learning
approaches. Multimed Tools Appl 82,
17521–17566 (2023).
https://doi.org/10.1007/s11042-02213797-w
Abhishek, Jindal, N. “Copy move and
splicing forgery detection using deep
convolution neural network, and
semantic segmentation”. Multimed
Tools Appl 80, 3571–3599 (2021).
https://doi.org/10.1007/s11042-02009816-3
Sharma, P., Kumar, M. & Sharma, H.
Comprehensive analyses of image
forgery detection methods from
traditional to deep learning
approaches: an evaluation. Multimed
Tools Appl 82, 18117–18150 (2023).
https://doi.org/10.1007/s11042-02213808-w
Image forgery
detection tools
A survey of image forgery
detection methods based on
deep learning.
How can deep learning
techniques improve the
detection and localization of
image forgeries?
The lack of a comprehensive and
reliable method to detect forged
certificates that are created using
image processing tools such as
Photoshop.
Deep convolution
neural network, and
semantic
segmentation
Detecting and localizing
image forgeries using deep
convolution neural network
and semantic segmentation.
How can deep convolution
neural network and semantic
segmentation be used to detect
and localize copy-move and
splicing image forgeries with
high accuracy and efficiency?
Lack of Improved hybrid algorithm
that uses color illumination, deep
convolution neural network, and
semantic segmentation to detect and
localize image forgeries
Photoshop, Corel
Paint Shop,
PhotoScape,
PhotoPlus, GIMP,
Pixelmator and GAN
approaches
Advanced image forensics
should incorporate more
advanced methodologies that
minimize executable time
and operational cost
What are the different types of
image forgery and how can they be
detected?
What are the challenges and
limitations of deep learning-based
image forgery detection methods?
What are the future directions and
open problems in image forgery
detection research?
The paper does not provide a
comprehensive comparison of the
performance, robustness, and limitations
of different image forgery detection
methods on various datasets & scenarios.
The paper does not address the challenges
posed by more advanced and realistic
image forgeries, such as deepfakes, face
swapping, and style transfer, which can
evade conventional and deep learningbased methods.
The paper does not explore the potential
of combining multiple modalities, such as
audio, video, and text, to enhance the
image forgery detection and localization
capabilities.
The paper does not investigate the ethical,
legal, and social implications of image
forgery detection and prevention, such as
privacy, consent, accountability, and trust.
- 14 -
Author(s) and Title
Methodology
(Algorithms)
Experiment Result
Research Questions
Research Gap
Ankit Katiyar and Dr. Arnav Bhavsar.
“Image Forgery Detection with
Interpretability” (2022).
https://arxiv.org/abs/2202.00908,
https://doi.org/10.48550/arXiv.2202.0
0908
Convolutional neural
network (CNN)
architecture
How can a learning-based
method using convolutional
neural network (CNN)
architecture detect copy-move
and inpainting-based image
forgeries with high accuracy
and interpretability?
The need for more effective and
reliable methods that can detect and
localize different types of image
forgeries, such as copy-move and
inpainting-based forgeries, with
high accuracy and interpretability
I. Psychoula, A. Gutmann, P. Mainali,
S. H. Lee, P. Dunphy and F.
Petitcolas, "Explainable Machine
Learning for Fraud Detection,"
in Computer, vol. 54, no. 10, pp. 4959, Oct. 2021, doi:
10.1109/MC.2021.3081249.
Explainability
methods
The proposed learning-based
method, which focuses on
the convolutional neural
network architecture, can
detect both copy-move
forgeries and inpaintingbased forgeries with high
accuracy and efficiency.
Nearest neighbor samples
produced the most
informative and consistent
explanations, while random
and stratified samples were
less reliable and more
variable.
The need for more effective and
reliable methods that can detect and
prevent different types of financial
fraud, such as credit card fraud,
insurance fraud, identity theft, and
money laundering, with high
accuracy and interpretability
Shaik, Cheman. “Preventing Forged
and Fabricated Academic Credentials
using Cryptography and QR Codes”.
International Journal of Computer
Science, Engineering and
Applications (IJCSEA), vol. 11, no. 1,
Feb. 2021, pp. 1-10,
Cryptography and QR
codes
How does the choice of background
dataset affect the explainability of
machine learning models for fraud
detection?
What are the trade-offs between
explainability and performance of
supervised and unsupervised models
for fraud detection?
How can explainability methods be
integrated into the fraud detection
pipeline to provide actionable
insights for decision makers?
How can cryptography and QR
codes be used to detect forgery and
fabrication in educational
certificates and academic
transcripts?
What are the requirements and
challenges of implementing the
proposed method in educational
institutes and universities?
How can the proposed method be
evaluated and compared with other
methods of verifying academic
credentials?
The paper does not present
any experimental results on
the proposed method of
preventing forged and
fabricated academic
credentials using
cryptography and QR codes.
- 15 -
The motivation for developing this
method of detecting forgery and
fabrication in educational certificates and
transcripts is not clearly stated
2.2.2. Gap Analysis
To identify a gaps in any literature review we have to consider the following points:
Lack of interpretability: Explainable AI models are designed to provide
transparency and interpretability to the decision-making process of AI systems.
However, the lack of interpretability in some models can make it difficult to
understand how they work and identify the reasons behind their decisions. This
can be a significant challenge in detecting forgery educational certificates.
Limited scope: Most existing techniques for detecting forgery educational
certificates are limited in scope and can only detect specific types of
forgery. This can be a significant limitation in detecting new and unseen types of
forgery.
Data quality: The quality of data used to train AI models can significantly
impact their performance. In the case of detecting forgery educational certificates,
the quality of data used to train the model can affect its ability to detect new and
unseen types of forgery.
Evaluation of techniques: The evaluation of existing and new techniques for
detecting forgery educational certificates is essential to determine their
effectiveness. However, the lack of standardized evaluation metrics and datasets
can make it challenging to compare the performance of different techniques.
Human factors: Human factors such as cognitive biases, social and cultural
norms, and ethical considerations can also impact the effectiveness of AI systems
in detecting forgery educational certificates. For example, human biases can
affect the quality of data used to train the model, leading to inaccurate results
There is a need for a technique that can efficiently and accurately detect the presence
of unseen forgeries in an image. Lack of Improved hybrid algorithm that uses color
illumination, deep convolution neural network, and semantic segmentation to detect
and localize image forgeries. The need for more effective and reliable methods that
can detect and localize different types of image forgeries, such as copy-move and
inpainting-based forgeries, with high accuracy and interpretability. The lack of
generalizability of existing certificate verification systems.
- 16 -
The motivation for developing this method of detecting forgery and fabrication in
educational certificates and transcripts is not clearly stated. The limitations of
conventional systems and the need for automated solutions. The research paper does
not mention any research gap. It is possible that the authors did not consider the need
for a research gap in their study. The paper does not provide a comprehensive
comparison of the performance, robustness, and limitations of different image forgery
detection methods on various datasets & scenarios. The paper does not address the
challenges posed by more advanced and realistic image forgeries, such as deepfakes,
face swapping, and style transfer, which can evade conventional and deep learningbased methods. The paper does not explore the potential of combining multiple
modalities, such as audio, video, and text, to enhance the image forgery detection and
localization capabilities. The paper does not investigate the ethical, legal, and social
implications of image forgery detection and prevention, such as privacy, consent,
accountability, and trust. The lack of a comprehensive and reliable method to detect
forged certificates that are created using image processing tools such as Photoshop.
- 17 -
CHAPTER THREE
3. Methodology
3.1. Research Methods
The methodology is the process of collecting data, analyzing it, selecting algorithms,
and interpreting it in order to achieve the goals and answer the research questions. It
is the science of studying how research is conducted methodically. We discussed
research methods for detecting forgery educational certificates. It involves a
combination of traditional and advanced methods to ensure the authenticity of
certificates. Some of the methodologies used in detecting forgery educational
certificates:
Visual Inspection: The first step in detecting forgery educational certificates is visual
inspection. This involves examining the certificate for any visible signs of forgery,
such as blurry or low-quality printing, misspellings, or inconsistencies in the design.
Authentication Markers: Educational institutions can include authentication markers,
such as holograms, watermarks, or security threads, on their certificates to make them
more difficult to forge. These markers can be difficult to replicate, and their presence
can indicate that the certificate is genuine.
Machine Learning Algorithms: Machine learning algorithms can be trained to detect
patterns and anomalies in educational certificates, making it easier to identify forgeries.
These algorithms can analyze various features, such as the font, color, and layout, to
determine the authenticity of a certificate.
Document Analysis: Document analysis is a scientific method used to examine the
authenticity of documents. This method involves analyzing the paper, ink, and other
materials used in the certificate to determine its authenticity.
Database Verification: Educational institutions can maintain a database of all issued
certificates. This database can be used to verify the authenticity of certificates by
comparing the certificate in question with the one in the database.
Third-Party Verification: Many institutions outsource the verification process to
third-party services. These services use a combination of traditional and advanced
methods to verify the authenticity of educational certificates.
By using a combination of these methodologies, educational institutions can ensure
the authenticity of their certificates and prevent fraudulent claims.
- 18 -
3.2. Research Design
Methodological Steps:
1. Data Collection: Gathering a large and diverse dataset of both genuine and
forged certificates is crucial. Different types of forgeries and various
university documents need to be represented.
2. Data Preprocessing: The dataset is cleaned, standardized, and formatted for
machine learning algorithms. This might involve image resizing, text
extraction, and feature engineering.
3. Model Training and Selection: Different ML models, such as convolutional
neural networks (CNNs) or recurrent neural networks (RNNs), are trained on
the prepared data. Their performance is evaluated and compared to select the
most accurate and efficient model.
4. Feature Extraction and Identification: The chosen model analyzes specific
features within the certificates, such as font inconsistencies, unusual word
patterns, paper grain, or pixel-level discrepancies, to identify signs of forgery.
5. Decision Making and Result Generation: The model outputs a prediction
based on the extracted features, classifying the certificate as genuine or
forged. This output can be presented with a confidence score indicating the
certainty of the prediction.
3.3. Materials
Explainable AI (XAI) plays a crucial role in detecting forged educational certificates
using artificial intelligence.
Black Box Problem: Traditional AI models, often referred to as "black boxes," can
achieve high accuracy in detecting forgeries. However, they lack transparency in their
decision-making process, making it difficult to understand how they arrive at their
conclusions.
Bias and fairness: If the training data contains biases, the model may inherit
them and unfairly target certain institutions or groups.
Trust and acceptance: Without understanding how the model works,
institutions and individuals may hesitate to rely on its outputs.
- 19 -
Debugging and improvement: Identifying and addressing false positives or
weaknesses in the model becomes challenging without insights into its
reasoning.
Detecting forged educational certificates involves two main types of materials:
1. Physical Materials:
Educational certificates: These can be paper-based or digital depending on
the issuing institution and timeframe. Different paper types, printing
techniques, and security features (holograms, watermarks) can be analyzed.
Reference documents: Genuine certificates from the issuing institution serve
as a baseline for comparison. Additional reference materials like historical
records, official templates, and security feature guidelines might be used.
Specialized equipment: Depending on the chosen detection methods, tools
like high-resolution scanners, microscopes, ultraviolet lamps, and forensic
analysis equipment might be necessary.
2. Data and Information:
Images of certificates: Both genuine and forged certificates are scanned or
photographed and processed as digital images for analysis.
Textual data: Extracted text from certificates, such as names, dates,
institution details and signatures, can be analyzed for inconsistencies or
unusual patterns.
Metadata: Information embedded within the document file, like timestamps,
author details, and software used, can be analyzed for discrepancies or clues to
forgery.
Databases: Institutional databases store issuing records and details of genuine
certificates, allowing for verification and data cross-checking.
Machine learning datasets: Large datasets of both genuine and forged
certificates are crucial for training and optimizing machine learning models
used for automated detection.
Research and technical documentation: Academic publications, guidelines
and technical reports on forgery detection techniques and technologies provide
valuable knowledge and reference material.
The specific materials used will depend on the chosen detection methods, available
resources, and the types of forgeries being targeted.
- 20 -
By effectively utilizing and managing both physical and digital materials, we can
develop more comprehensive and robust solutions for detecting forged educational
certificates and upholding the integrity of academic credentials.
3.4. Procedures
Detecting forged educational certificates requires a systematic procedure, typically
involving a combination of traditional methods and advanced technologies. The
breakdown of the key steps are:
1. Data Acquisition and Preparation:
Gather a diverse dataset: Collect genuine and forged certificates
representing various institutions, forgery techniques, and printing styles.
Preprocess the data: Standardize formats, extract relevant features
(text, images, metadata), and clean the data for accurate analysis.
2. Feature Analysis and Selection:
Identify key features: Analyze visual elements (fonts, signatures, paper
texture), textual patterns (inconsistencies, unusual formatting), and metadata
discrepancies to distinguish forged certificates.
Select informative features: Choose features that contribute most to accurate
forgery detection while considering efficiency and processing limitations.
3. Model Training and Optimization:
Choose a suitable ML model: Depending on the type of features and desired
accuracy,
select
an appropriate machine learning model, such as
Convolutional Neural Networks (CNNs) for image analysis or Recurrent
Neural Networks (RNNs) for textual data.
Train and refine the model: Train the model on the prepared data, evaluate
its performance, and adjust parameters or features for optimal accuracy and
generalization.
4. Forgery Detection and Verification:
Analyze individual certificates: Apply the trained model to new
certificates, extracting features and predicting the probability of forgery.
Combine with traditional methods: Integrate visual inspection, document
analysis, and database verification alongside the model's output for
comprehensive assessments.
5. Decision Making and Result Presentation:
- 21 -
Set a threshold: Determine a confidence level above which a certificate is
classified as forged based on the model's prediction score.
Present results: Report the forgery probability, highlight suspicious
features, and provide explanations (if using XAI techniques) for transparency
and trust.
6. Continuous Improvement and Monitoring:
Update the model: As new forgery techniques emerge, retrain the model on
expanded datasets to maintain its effectiveness.
Monitor performance: Track false positives and negatives, analyze
trends, and adapt the system to address evolving threats.
3.5. Evaluation
Evaluating the effectiveness of systems for detecting forged educational certificates is
crucial to ensure their accuracy, reliability, and fairness. Some key metrics and
approaches to consider:
Performance metrics:
Accuracy: The percentage of correctly classified certificates, both genuine and
forged.
Sensitivity: The rate of correctly identified forged certificates (True Positive Rate).
Specificity: The rate of correctly identified genuine certificates (True Negative Rate).
Precision: The proportion of identified forgeries that are truly forged (Positive
Predictive Value).
False Positive Rate (FPR): The rate of genuine certificates misclassified as forged.
False Negative Rate (FNR): The rate of forged certificates misclassified as genuine.
Evaluation methods:
Hold-out validation: Split the data into training, validation, and test sets and
evaluate the system's performance on the unseen test set.
K-fold cross-validation: Repeat the training and validation process multiple
times with different data partitions for a more robust evaluation.
AUC-ROC
curve:
Visualizes
the
trade-off
between
sensitivity
and
specificity, aiding in selecting an optimal operating point for the system.
Confusion matrix: Summarizes the classification results, facilitating analysis of
specific types of errors.
- 22 -
Human-in-the-loop evaluation: Combine automated detection with human
expertise for complex cases or in security-sensitive settings.
Challenges in evaluation:
Limited availability of ground truth data: Obtaining sufficient and diverse
datasets of both genuine and forged certificates can be difficult.
Evolving nature of forgery techniques: Systems need to adapt to detect new
and increasingly sophisticated methods of forgery.
Subjectivity and human error: Manual evaluation methods can be unreliable
and inconsistent.
3.6. Limitations
Detecting forgery in educational certificates can be a complex task, and there are
several limitations associated with the process. Some common limitations to consider
are the followings:
1. Evolving techniques: Forgers continuously improve their techniques to make their
forgeries more convincing. They may use advanced printing technologies, highquality materials, or even replicate security features accurately. As a result,
traditional detection methods may become less effective against sophisticated
forgeries.
2. Lack of standardized security features: Educational institutions may vary in the
security features they incorporate into their certificates. Some institutions may
have robust security measures, such as holograms, watermarks, or UV-reactive
inks, while others may have limited or no security features at all. This lack of
standardization makes it challenging to develop a universal detection method.
3. Availability of genuine certificates: Obtaining genuine certificates for reference
purposes can be difficult, especially if they are from different institutions or if the
issuing process is complicated. Without genuine certificates as a point of
comparison, it becomes harder to identify discrepancies in suspicious documents
accurately.
4. Limited access to advanced equipment: Some detection techniques, such as
advanced forensic analysis or specialized equipment, may require expertise and
resources that are not readily available to everyone. Access to tools like high-
- 23 -
resolution microscopes, UV light sources, or spectroscopy equipment may be
limited, making it challenging to conduct in-depth examinations.
5. Time and cost constraints: Thoroughly examining each certificate for potential
forgery can be a time-consuming process, particularly when dealing with a large
volume of documents. Additionally, employing experts or specialized services to
assist with forgery detection can be costly, making it impractical for some
institutions or organizations.
6. Human error and subjectivity: The process of detecting forgery in educational
certificates often relies on the judgment and expertise of individuals examining the
documents. However, human error and subjectivity can introduce biases or
overlook subtle signs of forgery, leading to false positives or false negatives.
3.7. Conceptual framework
A conceptual framework is a set of ideas, assumptions, and principles that guide the
design and implementation of a research project or a system. In the context of
detecting forgery in educational certificates, a conceptual framework may include the
following elements:
The problem statement: This defines the scope and motivation of the research
or system, such as the prevalence and impact of forged certificates, the
challenges and limitations of existing methods, and the objectives and expected
outcomes of the proposed solution.
The literature review: This provides a comprehensive and critical analysis of
the relevant existing works, such as the methods, techniques, tools, and
applications that have been used or proposed for detecting forgery in educational
certificates. This also identifies the gaps and limitations of the current state-ofthe-art and the opportunities for improvement and innovation.
The methodology: This describes the design and implementation of the
proposed solution, such as the data sources, the algorithms, the models, the
evaluation metrics, and the experiments that are used or developed for detecting
forgery in educational certificates. This also explains the rationale and
justification of the chosen methods and techniques and how they address the
research questions or objectives.
The results and discussion: This presents and analyzes the findings and
outcomes of the proposed solution, such as the performance, accuracy, efficiency,
- 24 -
and limitations of the methods and models for detecting forgery in educational
certificates. This also compares and contrasts the results with the existing works
and discusses the implications and contributions of the research or system.
The conclusion and future work: This summarizes the main points and
achievements of the research or system, such as the problem statement, the
literature review, the methodology, and the results and discussion. This also
provides recommendations and suggestions for future research or development,
such as the limitations, challenges, and open issues that need to be addressed or
explored further.
A conceptual framework for detecting forged educational certificates helps us
organize and understand the key components and processes involved in addressing
this complex challenge.
1. Inputs:
Physical/Digital Certificates: Both genuine and forged certificates in various
formats (paper, digital scans, images) serve as the raw data for analysis.
Supporting Data: Additional information like historical records, institutional
guidelines, security feature specifications, and known forgery techniques can
enrich the detection process.
Preprocessing
and
Standardization:
Raw
data
undergoes
cleaning, formatting, and feature extraction to prepare it for analysis.
2. Detection Techniques:
Traditional Methods: Visual inspection, document analysis, and database
verification remain vital initial steps, offering human expertise and direct
validation.
Machine Learning: Supervised and unsupervised learning algorithms trained on
large datasets of genuine and forged certificates can analyze textual, visual, and
metadata features to identify anomalies and inconsistencies.
Blockchain Technology: Securely storing and verifying certificate data on a
tamper-proof blockchain can enhance transparency and authenticity but might not
be directly involved in forgery detection.
3. Feature Analysis and Selection:
Visual Features: Font inconsistencies, paper properties, security element
analysis, and image-level texture comparisons can reveal signs of forgery.
- 25 -
Textual
Features:
Unusual
word
patterns,
formatting
irregularities, inconsistencies in dates and names, and extracted metadata can
provide clues.
Metadata Features: Embedding timestamps, author details, and software
information within the certificate file can be analyzed for discrepancies.
4. Decision Making and Output:
Model Classification: Based on extracted features and analysis, the system
classifies the certificate as genuine, forged, or suspicious requiring further
investigation.
Confidence Levels: Providing a confidence score alongside the classification
allows for nuanced judgments and understanding the certainty of the prediction.
Explainability and Transparency: Employing Explainable AI (XAI) techniques
can offer insights into the model's reasoning behind its decision, promoting trust
and understanding.
5. Feedback and Refinement:
Performance Monitoring: Evaluating the system's accuracy, sensitivity, and
other metrics on new data is crucial to identify weaknesses and areas for
improvement.
Adaptive Updating: Continuously updating the model with new data and
evolving forgery techniques ensures its effectiveness against emerging threats.
Human-in-the-Loop: Combining automated detection with human expertise can
address complex cases, refine interpretation, and maintain human oversight.
6. Societal Context and Considerations:
Ethical
Implications:
Addressing
potential
biases
in
data
and
algorithms, ensuring data privacy, and mitigating discriminatory outcomes are
essential ethical considerations.
Legal and Regulatory Frameworks: Clear legal frameworks for using detection
systems, defining forgery, and establishing consequences for illegal activities are
crucial.
Collaboration and Sharing: Fostering collaboration and data sharing among
institutions, researchers, and authorities can enhance detection capabilities and
combat global forgery challenges.
- 26 -
CHAPTER FOUR
4. Work and Budget Plan
4.1. Work plan
A work plan is a document that outlines the objectives, activities, timeline, and
expected outcomes of a project or a task. In the context of detecting forged
educational certificates, a work plan may include the following elements:
The background: This provides the context and motivation of the project or task,
such as the problem statement, the research questions, the literature review, and
the conceptual framework of detecting forged educational certificates.
The objectives: This defines the specific and measurable goals of the project or
task, such as the expected outputs, outcomes, and impacts of detecting forged
educational certificates.
The activities: This describes the steps and actions that will be taken to achieve
the objectives, such as the data collection, the data analysis, the method
development, the method evaluation, and the result dissemination of detecting
forged educational certificates.
The resources: This identifies the human, financial, technical, and logistical
resources that will be required or available for the project or task, such as the
team members, the budget, the equipment, the software, and the data sources of
detecting forged educational certificates.
The timeline: This specifies the duration and schedule of the project or task,
such as the start date, the end date, the milestones, and the deliverables of
detecting forged educational certificates.
The risks and challenges: This anticipates the potential difficulties and
uncertainties that may arise during the project or task, such as the data quality,
the data availability, the method validity, the method reliability, and the ethical
issues of detecting forged educational certificates. This also proposes the
mitigation strategies and contingency plans to address or overcome them.
The evaluation and monitoring: This defines the criteria and indicators that will
be used to measure the progress and performance of the project or task, such as
the quality, the accuracy, the efficiency, and the effectiveness of detecting forged
educational certificates. This also describes the methods and tools that will be
- 27 -
used to collect and analyze the data and feedback from the stakeholders and
beneficiaries of detecting forged educational certificates.
Table 1: Work Plan
No
Time Schedule
Activities
Week 1
Week 2
Week 3
Week 4
Week 5
Week 3
1 Literature review
Data collection and
2 preprocessing
3 Feature extraction
Neural network architecture
4 design and implementation
5 Training and validation
6 Post-processing
Real-world application
7 evaluation
8 Analysis and reporting
4.2. Budget plan
A budget plan is a document that outlines the objectives, activities, timeline, and
expected outcomes of a project or a task.
No
Items
Measurement
1 Pen
PCS
2 Dataset
Kb/Mb/Gb
3 Hardware and software
PCS
4 Travel and accommodation
Quantity
Cost / birr
15
Total cost
25
375
2
100000
200000
Trips
3
50
150
5 Publication fees
Pages
4
200
800
6 Miscellaneous expenses
Hours
20
5.75
115
Total Cost
201440
- 28 -
Data based
4.3. Future Work
Detecting forged educational certificates is an ongoing battle against ever-evolving
techniques. While current solutions are making strides, several promising avenues
offer exciting possibilities for future advancements:
Technological Advancements:
Deep learning advancements: Exploring new neural network architectures and
unsupervised learning techniques can potentially tackle complex forgeries and
adapt to unseen threats.
Transfer learning and domain adaptation: Transferring knowledge from other
domains (e.g., document forgery detection) can accelerate model training and
improve generalization for specific certificate types.
Hybrid approaches: Combining deep learning with traditional methods like
document analysis and physical property verification can improve overall
accuracy and reliability.
Biometric integration: Analyzing fingerprints, signatures, or other biometric
markers embedded in certificates could offer a strong layer of authentication.
Blockchain and decentralized solutions: Securely storing and verifying certificate
data on a tamper-proof blockchain can enhance transparency and prevent
centralized manipulation.
Focus on Explainability and Trust:
Advanced XAI techniques: Develop more comprehensive and user-friendly
explanations for AI decisions, allowing better understanding and trust in the
detection process.
Human-in-the-loop systems: Leverage AI for initial screening and detection, but
integrate human expertise for complex cases and ethical considerations.
Standardized reporting and guidelines: Establish transparent reporting protocols
for detected forgeries and consistent ethical guidelines for AI development and
deployment.
Collaboration and Data Sharing:
Global data pools and platforms: Building secure platforms for sharing data
among institutions and countries can significantly enhance detection capabilities
and combat global forgery rings.
- 29 -
Collaborative research and development: Fostering open research initiatives and
joint development efforts can accelerate innovation and optimize solutions for
diverse contexts.
Public awareness and education: Raising awareness about the risks of using
forged credentials and promoting responsible authentication practices among
individuals and institutions.
Addressing Societal Challenges:
Mitigating bias and discrimination: Developing fairness-aware algorithms and
data cleansing practices to prevent biased outcomes in detection, particularly
against specific institutions or groups.
Ensuring accessibility and affordability: Making advanced detection systems
accessible and affordable for institutions with limited resources, including those
in developing countries.
Legal and regulatory frameworks: Establishing clear legal frameworks for using
detection systems, defining forgery offenses, and outlining consequences for
illegal activities.
Future Directions:
Further Research: Investigating XAI in education remains an active area of
research. Researchers should explore novel techniques and evaluate their
effectiveness.
Practical Implementation: Applying XAI in real-world educational systems
requires collaboration between researchers, educators, and policymakers.
Hybrid Systems: Combining traditional techniques with automated methods can
offer a more robust and comprehensive approach to forgery detection.
Transfer Learning: Leveraging pre-trained models on large image datasets can
improve the efficiency and accuracy of forgery detection, even with limited
educational certificate data.
- 30 -
References:
Shaik, Cheman, Preventing Forged and Fabricated Academic Credentials using
Cryptography and QR Codes (February 28, 2021). International Journal of Computer
Science, Engineering and Applications (IJCSEA) Vol.11, No.1, February 2021,
Available at SSRN: https://ssrn.com/abstract=3831729
Abazi Chaushi, B., Selimi, B., Chaushi, A., & Apostolova, M. (2023). Explainable
Artificial Intelligence in Education: A Comprehensive Review. Communications in
Computer and Information Science, 1902, 48–71.
M. A. Ali ALAMERİ, B. Ciylan and B. Mahmood, "Computational Methods for Forgery
Detection in Printed Official Documents," 2022 ASU International Conference in Emerging
Technologies for Sustainability and Intelligent Systems (ICETSIS), Manama, Bahrain, 2022,
pp. 307-313, doi: 10.1109/ICETSIS55481.2022.9888875.
P. Yang, W. Fang, F. Zhang, L. Bai and Y. Gao, "Document Image Forgery Detection Based
on Deep Learning Models," 2022 International Symposium on Electrical, Electronics and
Information Engineering (ISEEIE), Chiang Mai, Thailand, 2022, pp. 36-41, doi:
10.1109/ISEEIE55684.2022.00014
Zanardelli, M., Guerrini, F., Leonardi, R. et al. Image forgery detection: a survey of recent
deep-learning approaches. Multimed Tools Appl 82, 17521–17566 (2023).
https://doi.org/10.1007/s11042-022-13797-w
Ali, Syed Sadaf, Iyyakutti Iyappan Ganapathi, Ngoc-Son Vu, Syed Danish Ali, Neetesh
Saxena, and Naoufel Werghi. 2022. "Image Forgery Detection Using Deep Learning by
Recompressing Images" Electronics 11, no. 3: 403.
https://doi.org/10.3390/electronics11030403
M. Patel, K. Rane, N. Jain, P. Mhatre and S. Jaswal, "Image Forgery Detection using
CNN," 2023 3rd International Conference on Intelligent Technologies (CONIT), Hubli, India,
2023, pp. 1-4, doi: 10.1109/CONIT59222.2023.10205377.
Kethepalli Mallikarjuna, Pushan Kumar Dutta, N. Sateesh Kedarnath Kumar. “Detecting Forged
Scan of Educational Certificates Using a New Feature Set Matching Algorithm”. International
Journal of Tomography & Simulation, 2019.
http://www.ceser.in/ceserp/index.php/ijts/article/view/5899
C. N. Bharti and P. Tandel, "A survey of image forgery detection techniques," 2016 International
Conference on Wireless Communications, Signal Processing and Networking (WiSPNET),
Chennai, India, 2016, pp. 877-881, doi: 10.1109/WiSPNET.2016.7566257.
Zanardelli, M., Guerrini, F., Leonardi, R. et al. Image forgery detection: a survey of recent deeplearning approaches. Multimed Tools Appl 82, 17521–17566 (2023).
https://doi.org/10.1007/s11042-022-13797-w
Sharma, P., Kumar, M. & Sharma, H. “Comprehensive analyses of image forgery detection
methods from traditional to deep learning approaches: an evaluation”. Multimed Tools Appl 82,
18117–18150 (2023). https://doi.org/10.1007/s11042-022-13808-w
Ankit Katiyar and Dr. Arnav Bhavsar. Indian Institute of Technology Mandi, HP, INDIA.
“Image Forgery Detection with Interpretability” (2022). https://arxiv.org/abs/2202.00908,
https://doi.org/10.48550/arXiv.2202.00908
- 31 -