Year: 2025
Type: dissertation
Abstract: The usage of deepfake technology has increased rapidly over the years, posing serious threats to businesses and people who are affected. It facilitates activities such as identity fraud, social engineering attacks, and misinformation, causing loss of reputation. This research work proposed a hybrid Deep Learning model that integrates Convolutional Neural Network, EfficientNet, and Vision Transformers to detect deepfake images. A custom dataset that includes a partial amount of the benchmarked public dataset (FaceForensics++, DFDC, and Celeb-DF-V2) and images from Google searches to cover the latest manipulated images is used to train the model. Hybrid architecture was validated against the established detection models, demonstrating an accuracy up to 93% and an Area Under the Curve (AUC) of 97.81%. Strengthening detection models and addressing challenges related to datasets like bias and high-quality data forgery were the focus of this work. This research work contributes to building more robust cybersecurity measures with the help of Artificial Intelligence to safeguard digital media integrity.
Author: Nagaraj Kumar, Sharan
Institution: National College of Ireland
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7165232746
