This work offers a copy-move tamper detection algorithm based on the enhanced DeepLabv3+ to address the issues of difficult extraction of image tampering edge features and low positioning accuracy of the tamper region.In the enhanced DeepLabv3+,a dual attention mechanism module is introduced to capture contextual information to improve the adaptability of the model to the tampering region,the residual refinement module used to optimize the prediction mask to enhance the sensitivity of the model to the tampering boundary,and a new hybrid loss function performed to help better model training of the mapping relationship between tampered images and corresponding real masks at the pixel level and image level.The experimental results show that the three evaluation indexes of the enhanced DeepLabv3+ on the COPYMOVE_NIST and COPYMOVE_COCO datasets are higher than those of FCN,U-Net and DeepLabv3+,with the detection accuracy standing at 0.929 and 0.895 respectively,which verifies that the enhanced detection algorithm of DeepLabv3+ can effectively extract the edge features of image tampering and solve the problems of missed edge pixel detection and false detection.
NETL
NSTL
万方数据
