Enhancing deepfake detection generalization through component-based development in a web platform

Koh, Yeong Keong (2025) Enhancing deepfake detection generalization through component-based development in a web platform. Final Year Project, UTAR.

PDF Download (2963Kb)

Abstract

This report outlines the design, development, and evaluation of a deepfake detection system aimed at providing an accessible and scalable solution for detecting manipulated media. The system leverages advanced machine learning models, including both single-model and ensemble-based detection methods, to identify deepfakes in images. The platform supports easy image uploads, efficient model processing, and reliable result presentation, offering users the ability to choose between various detection models based on their needs. Key features include user authentication and role management, image validation, preprocessing, and real-time inference with confidence scores. The system utilizes a modular architecture to integrate new models seamlessly, ensuring scalability and maintainability. Performance benchmarks are met, including a processing time of less than 800ms per image and a 99.9% uptime for system reliability. The accuracy of the ensemble detection method is validated through extensive testing on benchmark datasets, achieving a high F1-score. This project addresses the growing concern of deepfake threats in digital media and aims to provide an easy-to-use, robust tool for both non-technical users and advanced administrators. The system is designed with a focus on usability, accuracy, performance, and security, ensuring it meets the challenges posed by modern deepfake detection. Keywords: Machine Learning, Generative AI, Deepfake, Component-based, Image Classification, Artificial Intelligence, AI Generalization Subject Area: QA75.5-76.95 Computer Science

Actions (login required)