TruthLens
Final Year Project Proposal Session
2022-2026
A project submitted in partial fulfilment of the COMSATS University
Degree of BS in Computer Science / Software Engineering (CUI)
Group Members
Irfan Karim (SP22-BCS-036)
Aliyan Ali (SP22-BCS-035)
Sohail Ahmad (SP22-BSE-154)
Department of Computer Science COMSATS University Islamabad,
Abbottabad Campus
18 March 2025
Title of the Project: TruthLens
AI-Based Deepfake Image Detection for Web and Mobile
Final Year Project Proposal
Session 2025-2026
Project Registration
Type of Project:
☐ Traditional
☐ Industrial
☐ Continuing
☑ Research & Development
Nature of Project:
☑ Development
☑ Research & Development
Sustainable Development Goals (SDGs):
☑ Industry, Innovation, and Infrastructure
☑ Decent Work and Economic Growth
☑ Quality Education
☑ Cybersecurity & Misinformation Prevention
Area of Specialization:
☑ Artificial Intelligence (AI)
☑ Data Science and Analytics
☑ Web Development
☑ Mobile App Development
Project Abstract
Deepfake images emerged as significant issues regarding misinformation and identity fraud and
online scams while spreading in recent years. An AI-based deepfake image detection system
forms the core of this project because it establishes a method to evaluate image authenticity.
Deep learning CNNs together with Transformer-based architectures power the technology that
identifies altered media.
The project will include
Users can access a web-based and mobile application to load and confirm image content.
Real-time detection functionality will be available through the Flutter mobile application
design.
The system implements an AI processing model that receives deepfake and real image datasets
for its extensive training.
Fast scalable processing capability will be enabled through the implementation of cloud service
integration.
Our goal focuses on delivering a simple yet AI-powered deepfake detection software to help
users check facts along with protecting cybersecurity and blocking false information.
Introduction
All advancements in deepfake technology result from AI developments which create highly
authentic fakes that hinder people's ability to detect true or fake content. The deployment of
deepfake images disrupts traditional information security measures and damages public
content reliability as well as compromising user privacy protections. The proposed AI-based
deepfake detection solution analyzes image pixels to detect anomalies and evaluate visual
artifacts and validates metadata consistency to fulfill the set project objectives.
The main objectives of this project are:
To develop an AI model that identifies deepfake images with outmost accuracy.
To create a web and mobile platform for fake content detection.
We require cloud-based AI inference as the basis for running quick and scalable
analytical operations.
Examining all kinds of manipulated content and synthetic media face swaps is included
in this analysis.
Success Criterion
For this project to be considered successful, the system must:
1. Achieve an accuracy of 80% in fake ai generated image detection.
2. Provide real-time image analysis results .
3. Have a user-friendly web and mobile interface .
4. capable of analyzing 100 images per day through cloud-based processing.
5. The system needs to show its ability to detect multiple variations of deepfake
generation methods.
Related Work
Several deepfake detection models exist, such as:
The Facebook-AI researcher collaboration has produced DeepFake Detection Challenge
(DFDC) models as part of their work in this field.
FaceForensics++ serves as a vital tool to train detection models for the market since
users consider it the most suitable choice.
The Microsoft Image Authenticator tool in the system conducts examinations of small
disturbances contained in images.
The Hugging Face Transformers system supports AI content classification through its
models.
Our initiative bases its core operation on modern detection techniques through the deployment
of time-sensitive state-of-the-art AI modeling capabilities.
Project Rationale
Deepfakes serve as tools for creating fraudulent news together with stealing identities and
carrying out monetary fraud and generating cybersecurity danger situations. Our objective is
to develop a verification system based on AI which lets users check online image contents
before making trust decisions. The solution provides essential protection against false
information to journalists along with cybersecurity experts and regular individuals.
Aims and Objectives
A deep learning-based artificial intelligence system needs creation to find deepfake
images.
The project will produce a real-time verification system which consists of a web platform
developed through React.js technology and mobile application built with Flutter.
The system should receive training from datasets which include FaceForensics++, DFDC
and Celeb-DF.
The system requires integration of Google Cloud AI services or AWS AI services to scale
the detection capabilities.
The system design should enable accurate processing operations with minimal time
requirements to guarantee user-friendly performance.
Scope of the Project
This project includes:
We develop the deepfake detection artificial intelligence through training procedures.
An interactive user interface will be developed through web application development
processes to enable users for both image uploading and image checking functions.
Mobile App Development: Build a Flutter-based app for mobile verification.
The system utilizes cloud infrastructure to deploy AI analytical services which run in real-
time.
Performance Testing: Ensure accuracy, efficiency, and scalability.
Proposed Methodology and Architecture
1. Dataset Collection – Gather ai generated and real images datasets.
2. Model Training – Train CNNs and Transformers for fake content detection.
3. Web & Mobile Development – Build React.js (web) and Flutter (mobile) interfaces.
4. Cloud Integration – The company should utilize Google Cloud AI or AWS for scalable
processing needs.
5. Testing & Deployment – Evaluate performance and deploy the system.
Constraints
1. High Computational Requirements – Training deep learning models demands high-
performance GPUs for their successful operation.
2. Dataset Limitations – Availability of high quality datasets for training and testing.
3. False Positives/Negatives – AI models sometimes misclassify images in some senarios.
4. Evolving Deepfake Techniques – such system requires a continuous improvement
approach.
Stakeholders
1. End Users – Individuals, and cybersecurity professionals for the verification of images.
2. AI Researchers – Experts analyzing and improving fake content detection techniques.
3. Government Agencies – Organizations for checking misinformation and fraud.
4. Tech Companies – Deepfake detection functions have been integrated into security
mechanisms of various platforms.
5. Media & News Outlets – Journalists need a fact-checking system.
Tools and Technologies
Component Technology
AI Framework TensorFlow, PyTorch
Web Development React.js, Flask
Mobile Development Flutter
Database Firebase, PostgreSQL
Cloud Services Google Cloud AI, AWS
Dataset FaceForensics++, DFDC, Celeb-DF
The project establishes an efficient deepfake detection system based on artificial intelligence
which operates on a scalable framework. Research and development resulting from this
partnership between AI and cloud computing with web/mobile applications brings forth
significant impact.
