Edge-aware Regional Message Passing Controller for Image Forgery Localization

Digital image authenticity has promoted research on image forgery localization. Although deep learning-based methods achieve remarkable progress, most of them usually suffer from severe feature coupling between the forged and authentic regions. In this work, we propose a two-step Edge-aware Regional Message Passing Controlling strategy to address the above issue. Specifically, the first step is to account for fully exploiting the edge information. It consists of two core designs: context-enhanced graph construction and threshold-adaptive differentiable binarization edge algorithm. The former assembles the global semantic information to distinguish the features between the forged and authentic regions, while the latter stands on the output of the former to provide the learnable edges. In the second step, guided by the learnable edges, a region message passing controller is devised to weaken the message passing between the forged and authentic regions. In this way, our ERMPC is capable of explicitly modeling the inconsistency between the forged and authentic regions and enabling it to perform well on refined forged images. Extensive experiments on several challenging benchmarks show that our method is superior to state-of-the-art image forgery localization methods qualitatively and quantitatively.

1. Introduction

The forged or manipulated images pose risks in various fields, such as removing copyright watermarks, generating fake news, and even falsifying evidence in court [32]. The growing technology of forgery will cause a crisis of trust and affect social equity. Therefore, the detection of image forgery is of great significance. The crucial aspect of the detection is to model the inconsistency between the forged and authentic regions and to locate the forged regions on the suspicious image, i.e., image forgery localization (IFL). However, as postprocessing techniques such as GAN [16], [26], [63], VAE [27], [44] and homogeneous manipulation [7], [33] are wildly utilized, images can be easily tampered in a visually imperceptible way. These techniques constantly aim to couple the forged and authentic regions' features, making image forgery localization challenging. Therefore, in order to accurately locate the image forgery region, it is particularly essential to decouple the features between forged and authentic regions. Figure 1.

The difference between previous methods and ours. Our method controls the message passing between the forged and authentic regions, ignored by the previous method. As shown by X.