Abstract: In this paper, we construct CAMU-Net, an image forgery detection method, to obtain evidence of copy-move forgery areas in images. In CAMU-Net, the hierarchical feature extraction stage (HFE_Stage) is used to extract multi-scale key feature maps. Next, a hierarchical feature matching stage (HFM_Stage) based on self-correlation combined with a multi-scale structure is designed to predict copy-move forgery areas with different scales of information. To optimize the matching results, we design a coordinate attention-based resource allocation stage (CARA_Stage), which uses a location and channel attention mechanism to assign more weight to copymove areas. In this way, useful information can be strengthened while irrelevant information is suppressed. To effectively use the multi-scale prediction results in the multi-scale feature fusion-based up-sampling stage (MFFU_Stage), we integrate the high-level and low-level information into one information flow. By combining the global feature information of the deep layers and the location details of the shallow layers, the performance of CMFD can be improved. To demonstrate the validity of our model, we compare it with a variety of traditional methods and deep learning methods. The results show that our performance is outstanding. In particular, on the COVERAGE dataset, our AUC is 87.3%, which is 2.4% higher than the second place. In addition, we design a variety of baseline methods to perform several ablation experiments to demonstrate the validity of the modules in this model.