Image tampering poses a significant threat to the authenticity of visual information, with potential ramifications spanning misinformation propagation, financial fraud, and identity forgery. Existing detection approaches leverage frequency-domain features to uncover manipulation artifacts imperceptible to the human eye. However, these features are often inadequately integrated with RGB representations, limiting detection accuracy for subtle manipulations. To address this, we propose a high-resolution Transformer-based framework that incorporates a Tampering Artifact Enhancement Module (TAEM) and a Transformer decoder equipped with multi-scale deformable attention and mask attention mechanisms. The TAEM aggregates forensic artifacts extracted via DCT, SRM, and BayarConv filters to enhance sensitivity to tampering clues. The decoder enables effective cross-scale feature interaction and refines tampering masks in a coarse-to-fine manner. Extensive experiments on five publicly available forensics datasets demonstrate that our method achieves superior performance, with an average F1 score of 61.7%, AUC of 88.5%, and IoU of 56.0%, consistently surpassing state-of-the-art methods. These results highlight the proposed model’s strong generalization ability and practical value for real-world forensic analysis.