Recent Advances in Copy–Move Forgery Detection: A Deep Learning Oriented Survey

Abstract

With the widespread use of digital images in news reporting, forensic analysis, and social media, the problem of image tampering has become increasingly prominent. To address this, researchers have proposed various Copy–Move Forgery Detection (CMFD) approaches, primarily based on block segmentation, key points, and deep learning. However, with the rapid development of emerging technologies such as Transformers, Generative Adversarial Networks (GANs), and Diffusion Models, while exploratory research has emerged in image tampering detection, a systematic review of these approaches is lacking. Therefore, this article systematically reviews the development of CMFD within a three-part framework, focusing on the latest progress from 2023 to 2024. Specifically, this article reviews the classic contributions and evolution of block-based methods and analyzes the application value of key point-based methods in low-computing scenarios such as mobile devices and remote sensing. It also explores the advantages of deep learning methods in detection accuracy, robustness, and adaptability to AIGC scenarios. This paper further compares the characteristics and application scenarios of three types of methods (Transformer, GAN and Diffusion Models), and further proposes future development directions.