Title: Exposing deepfakes in online communication: detection based on ensemble strategy

Authors: Jie Xu; Guoqiang Wang; Tianxiong Zhou

Addresses: Changzhou Meteorological Bureau, Changzhou, 213125, Jiangsu, China ' Changzhou Meteorological Bureau, Changzhou, 213125, Jiangsu, China ' Changzhou Meteorological Bureau, Changzhou, 213125, Jiangsu, China

Abstract: In recent years, deepfake techniques have appeared in people's lives. As a product of deep learning, it can generate realistic face-swapping videos. Due to its high fidelity, deepfake is often used to produce porn videos and guide the public opinion, so as to pose a great threat to social stability. Previous studies have been able to improve detection accuracy. This paper aims to improve the detection ability of existing schemes by using the ensemble learning scheme from the perspective of model learning. Specifically, our scheme includes feature extraction, feature selection, feature classification, and a combination strategy. The experimental results on several datasets demonstrate that our scheme can effectively improve the detection ability of the model.

Keywords: deepfake detection; ensemble strategy; online communication; video forensics; deep learning.

DOI: 10.1504/IJAACS.2024.135935

International Journal of Autonomous and Adaptive Communications Systems, 2024 Vol.17 No.1, pp.24 - 38

Received: 17 Nov 2021
Accepted: 16 Dec 2021
Published online: 10 Jan 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article