融合注意力与图卷积的嵌套U型图像鉴伪网络OA

Image tampering poses a serious threat to digital media security,creating an urgent need for more robust detection methods.Exist-ing forgery detection approaches often struggle to effectively capture fine-grained tampering features and fuse multi-scale contextual information.To address these issues,we adapt a nested U-shaped network from the saliency detection domain to the image forensics task.In view of the subtle differences between tampered and authentic regions,we propose,through systematic experiments,a detection network that combines multi-scale attention with graph convolution.Experimental results on four benchmark datasets demonstrate that the proposed method achieves superior performance compared to current mainstream deep learning approaches.