Image Tampering Detection Approaches Using Multi-Stream Information Fusion and Transformer Architecture

Abstract

With the development of technology, image editing tools are becoming increasingly powerful and image tampering is also becoming increasingly easy. A large number of manipulated images are spreading on social media, causing certain damage to social stability. At the same time, it also has a certain impact on the law and the news media. The goal of image forgery detection is to detect and locate the altered parts within digital images. Thus protecting the authenticity of the image. This article provides a brief introduction to image tamper detection from the perspective of deep learning. First, sort out commonly used public datasets in the field to provide a foundation for method validation; Second, focus on elaborating methods based on multi-stream information fusion and those based on Transformer and other network architectures, respectively introducing classic models and the latest research progress of each type of method. Finally, conduct a multidimensional detailed comparison of the two core methods and analyze their advantages.