Title: A lightweight deep learning ensemble model based on convolutional neural network for deepfake detection in videos
Authors: Richa Gupta; Varun Borar; Nishika Khatri; Rakesh Garg; Neetu Singla; Ritvik Garg
Addresses: Department of Computer Science and Engineering, Amity University Uttar Pradesh, Noida, India ' Department of Computer Science and Engineering, Amity University Uttar Pradesh, Noida, India ' Department of Computer Science and Engineering, Amity University Uttar Pradesh, Noida, India ' Department of Computer Science and Engineering, Gurugram University, Gurugram, Haryana, India ' Department of Computer Science and Engineering, Gurugram University, Gurugram, Haryana, India ' Department of Computer and Communication Engineering, Manipal University, Jaipur, Rajasthan, India
Abstract: The advancement in Artificial Intelligence has promoted significant growth in developing highly realistic forged and manipulated media, commonly known as deepfakes. As a result, deepfakes have been associated with several positive applications in the education and entertainment industries. However, at the same time, deepfakes have the potential to be part of criminal activities, such as identity theft, pornography, or spreading false information resulting in distrust in digital media. As technology continuously evolves, it is crucial to ensure the development of techniques or algorithms to prevent unintended consequences. A lightweight and computationally efficient convolutional neural network (CNN) model, NVNet, to detect media forgery is proposed. The CNN model is trained on a dataset extracted from FaceForensic++. The dataset consists of four facial forgeries: i.e., Deepfakes, Face2Face, FaceSwap, NeuralTexture and original sequences. At last, we propose an ensemble model trained on different datasets to detect deepfakes.
Keywords: deepfakes detection; synthetic media; artificial intelligence; convolutional neural network; CNN; detection techniques; ensemble model.
DOI: 10.1504/IJQET.2024.143752
International Journal of Quality Engineering and Technology, 2024 Vol.10 No.3, pp.248 - 264
Received: 21 Apr 2024
Accepted: 05 Sep 2024
Published online: 06 Jan 2025 *