| Abstract: | The rapid advancement of deepfake technology poses significant challenges to the authenticity and
integrity of digital media, leading to widespread concerns in various sectors, including politics,
media, and personal relationships. This study aims to develop an AI-enhanced Convolutional
Neural Network (CNN) model for detecting deepfake images, addressing the limitations of
traditional detection methods that struggle against sophisticated manipulation techniques. By
leveraging state-of-the-art deep learning architectures, specifically the XceptionNet model, this
research explores the efficacy of advanced feature extraction techniques and data augmentation
strategies to improve detection accuracy.
The proposed system utilizes the FaceForensics++ dataset, which includes both authentic and
manipulated images, to train and evaluate the model. Experimental results demonstrate that the
AI-enhanced CNN model significantly outperforms traditional approaches, achieving a binary
classification accuracy of 86.91% and a type classification accuracy of 70.50%. These findings
indicate that the model effectively identifies subtle artifacts and inconsistencies that are
characteristics of deepfake manipulations.
This research not only contributes to the field of digital forensics but also emphasizes the need for
ongoing advancements in detection methodologies to combat the evolving landscape of deepfake
technology. Future work will focus on expanding the dataset, enhancing real-time detection
capabilities, and integrating interdisciplinary approaches to address the broader societal
implications of deepfakes. Ultimately, this study aims to empower individuals and organizations
with reliable tools to discern authentic media from manipulated content, fostering a safer and more
trustworthy digital environment. |
