Deepfake Detection in Images and Videos: A Comprehensive Review of Techniques and Challenges
Year: 2026
Type: conference-paper
Abstract: Deepfakes are artificially generated or manipulated multimedia contents created using advanced deep learning techniques, posing serious threats to digital security, privacy, and trust. The rapid advancement of generative models such as Generative Adversarial Networks (GANs) and autoencoders has significantly improved the realism of deepfake images and videos, making their detection increasingly challenging. This survey provides a comprehensive review of deepfake detection techniques across both image and video modalities. The study categorizes existing approaches into image-based, video-based, graph-based, hybrid, and lightweight detection methods, highlighting their key characteristics and limitations. It also presents an overview of widely used benchmark datasets, including Celeb-DF, DFDC, DeeperForensics, and ForgeryNet, emphasizing their role in evaluating model robustness and generalization. In addition, the survey highlights the role of machine learning, deep learning, and transformer-based architectures in providing improved detection performance. Furthermore, we also present a comparison of the performance of various models across all 5 datasets in terms of accuracy, complexity and cross-dataset generalization. The limitations of datasets, overwhelming processing costs and the fact that highly realistic deepfakes could be undetectable are among key challenges they identify. Eventually, some future research direction is presented for achieving the robust, efficient, and real-time systems for deepfake detection.
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7165638457