Object-level Copy-Move Forgery Image Detection based on Inconsistency Mining

TL;DR

This work tackles copy-move forgery detection with post-processing challenges that obscure traces and hinder object-level integrity. It introduces IMNet, which uses customized source-region prototypes (SRP) and tampered-region prototypes (TRP) combined with an Inconsistency Mining module (comprising Local Correspondence Correlation Detection and Reverse Gate Mechanism) to iteratively refine object-level representations. The key technical contribution is the inconsistency-based refinement, including the update rule $F^{'}_{z} = F_z \otimes (1-\mathrm{Gate}(F_z))$, and the use of prototype updates $s+1$ and $t+1$ to converge toward accurate detection masks. Experiments on USC-ISI CMFD, CASIA v2.0, and CoMoFoD demonstrate superior accuracy and robustness, with ablations showing the value of four prototype updates per iteration and channel-aware inconsistency for improved performance.

Abstract

In copy-move tampering operations, perpetrators often employ techniques, such as blurring, to conceal tampering traces, posing significant challenges to the detection of object-level targets with intact structures. Focus on these challenges, this paper proposes an Object-level Copy-Move Forgery Image Detection based on Inconsistency Mining (IMNet). To obtain complete object-level targets, we customize prototypes for both the source and tampered regions and dynamically update them. Additionally, we extract inconsistent regions between coarse similar regions obtained through self-correlation calculations and regions composed of prototypes. The detected inconsistent regions are used as supplements to coarse similar regions to refine pixel-level detection. We operate experiments on three public datasets which validate the effectiveness and the robustness of the proposed IMNet.

Figures (2)

• Figure 1: The input of the IMNet is the copy-move forgery image. The suspect region is obtained to refine the detection effect by mining the inconsistency of a pair of custom prototypes and the coarse similar regions.
• Figure 2: Comparison of detection performance of different algorithms. The first column is to copy-move forgery images, the second column is groundtruth, and the third to ninth columns are mask generated different methods