Splicing Tampered Image Detection Algorithm Fused With Multi-Scale Features
In view of the problem that the current detection and localization methods of splicing tampered images mainly focus on detecting and locating a small range of tampered regions,while the model performance of target objects at uneven size positions is poor,a new network architecture DEUNet for the detection and localization of stitched tampered images is proposed.DEUNet introduces efficient additive attention and bidirectional residual blocks on the basis of UNet to deal with features at different scales,which can reduce the complexity of the model while locating the large-scale tampering region more completely,and combine cross-entropy and Dice loss function to better balance the classification accuracy and segmentation accuracy.Experimental results show that the pro-posed method has better performance than other algorithms and has good robustness.In conclusion,DEUNet suc-cessfully solves the challenge of unfixed size position targets and the experimental verification performance is bet-ter than that of the latest algorithms.
NETL
NSTL
万方数据
