Deepfake Detection in Facial Images Using Convolutional Neural Networks
Atharva Jagam, Nomesh Patel, Bharathi Chidirala, Bibhudendra Acharya · 2025
Artificial intelligence achievements in synthetic media synthesis have significantly enhanced the production of ultrarealistic deepfake images that cannot be distinguished from originals. The advancement raises significant concerns about misinformation and invasion of privacy because access to convincingly proficient deepfakes is slowly becoming available. This paper introduces a new approach to detecting deepfakes through the creation of a custom dataset comprising 579,851 images. The dataset is constructed by merging the OpenForensics: Multi-Face Forgery Detection and Segmentation In-TheWild dataset, which contains manipulated images, with a set of AI-generated images. This combination is designed to address recent advancements in AI-generated content. The proposed model then utilizes an eight-layer Convolutional Neural Network (CNN) architecture. The model achieves an accuracy of 84.41%, with a loss value of 0.3484, demonstrating its effectiveness in distinguishing between authentic and AI-generated images. Additionally, a comparative analysis is performed between the proposed model and the CIFAKE image classification model, evaluating both models in terms of performance and efficiency. The analysis highlights the strengths of the proposed solution in detecting AI-generated forgeries. The datasets used in this study, including the OpenForensics dataset and the AI-generated images, are accessible through open-source platforms. AI-generated images were also created using state-of-the-art Stable Diffusion models, ensuring the proposed solution remains adaptable to evolving AI-generation techniques.