基于ConvNeXt-Mamba的双编码器图像伪造检测OA

Image forgery detection is a fundamental and critical task in the field of cybersecurity.Convolutional neural network(CNN)is the mainstream approach in image forgery detection.However,CNN typically extracts only local fea-tures,making it difficult to capture global characteristics.To address this limitation,this study proposes a dual-encoder architecture integrating Mamba and ConvNeXt,where Mamba is responsible for capturing global contextual features,while ConvNeXt focuses on local detail features.The synergy between these two components enables comprehensive fea-ture extraction.To further enhance the representation of key features,a channel attention module(SE block)is introduced,which adaptively adjusts the weights of feature channels to improve feature expressiveness.To mitigate the issue of missed detections caused by complex forged region boundaries,an edge loss term is incorporated to enhance the model's accuracy in identifying forgery contours.Experiments conducted on four benchmark datasets,including CASIAv1,demon-strate that the proposed method achieves an average improvement of 0.015 in AUC(area under the curve)and 0.054 in F1-score,significantly outperforming existing approaches.Notably,it exhibits superior robustness in handling complex arti-facts and blurry boundary scenarios.