ADVANCEMENTS IN DEEP LEARNING FOR IMAGE FORGERY DETECTION: FROM COPY-MOVE TO DEEPFAKE MANIPULATIONS

Abstract

In recent years, the ease of image manipulation has led to the widespread creation and distribution of doctored and altered photos via the Internet and media. Various methods have been proposed to identify whether an image is authentic and, in some cases, to detect specific manipulations or fabrications. This study offers an in-depth review of current image forgery detection techniques using Deep Learning (DL), with a particular focus on methods that address copy-move and splicing attacks. Additionally, the study explores the use of DeepFake-generated content in image manipulation. Given the dominance of deep learning-based techniques in the field, which have shown excellent performance across various benchmark datasets, this paper comes at a crucial time. We describe the methods, the datasets employed for training and validation, and highlight their key features. A comparison of their performance is also provided, followed by a discussion on potential future research directions in deep learning architecture, evaluation approaches, and the creation of datasets to facilitate easier comparison of techniques.