JPEG artifacts produced during the JPEG compression may obscure forgery artifacts, impairing the efficiency of regular forgery detection and localization approaches. To address the issue, we introduce a Multi-Scale Contextual Spatial-Channel Correlation Network, which has been designed for detecting and locating forgeries. Our MSCSCC-Net uses multi-scale mechanisms, which improves forgery detection and localization performance by better handling the scale variation of the forged areas and further helps to remove JPEG artifacts from forgery artifacts at different scales. We design a contextual spatial correlation module (CSCM) and a contextual channel correlation module (CCCM) to generate contextual spatial-channel features at different scales, with the aim of distinguishing between JPEG artifacts, forgery artifacts, and authentic regions. Finally, use the fused features to detect forgery and generate a predicted mask in a coarse-to-fine progression. Besides, we also use the fused features to generate a restored image and make the removal of JPEG artifacts a task of our network, ensuring the retention of forgery artifacts during the removal process. Extensive experimental results demonstrate our MSCSCC Net's advantages over the most advanced approach in forgery detection and localization of JPEG compressed images.
