Year: 2025
Type: article
Abstract: The rapid advancement of Deepfake technology, powered by deep learning algorithms and Generative Adversarial Networks (GANs), presents a paradigm shift in digital content creation and manipulation. This technology, capable of generating highly realistic but entirely synthetic audiovisual content, has implications that stretch across various domains, from entertainment to politics, posing both opportunities for innovation and risks for misinformation and privacy violations. This paper provides a comprehensive overview of the evolution of Deepfake technology, highlighting key developments in AI- driven synthetic media creation. It delves into the state-of-the-art detection techniques that leverage both appearance-based and geometric features to combat the proliferation of Deepfakes, emphasizing the importance of precise geometric analysis and temporal modelling in enhancing detection robustness, especially in the face of sophisticated manipulation techniques that evade traditional detection methods. Through an analysis of contemporary datasets and detection frameworks, the paper assesses the effectiveness and limitations of current approaches, underscoring the challenges posed by video compression and digital noise in real-world scenarios. Furthermore, it discusses the profound societal impact of Deepfakes, from the erosion of trust in digital media to legal and ethical dilemmas, and proposes future directions for both technological advancements in Deepfake generation and detection, and policy measures to mitigate their adverse effects. By bridging the gap between technological capabilities and ethical considerations, the research aims to foster a deeper understanding of Deepfakes’ dual potential to both enrich and deceive, calling for a balanced approach to harnessing and regulating this powerful technology.
Authors: Upasana Bisht, Pooja
Institution: Guru Gobind Singh Indraprastha University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4408299839