Image Tampering Detection Algorithm Based On Deep Learning

This thesis presents an accurate,reliable,and robust manipulation network and tamper detection method based entirely on deep-learning algorithms,and a new approach for copy move images tampers detection and manipulation network for image forgery.The highly efficient detection techniques of depth learning are applied instead of conventional digital images forensic.First,we propose counter-image forgery,which usually requires different forms of operations and combinations,and a deep,unified neural architecture is proposed.Compared with other existing methods,the operation is an end-to-end network,which detects and locates without unnecessary pre/post processing.The manipulation network is a completely convergent network,which manages images of any size and several known forms of forgery,namely,splicing,copying,eliminating,enhancing,or unknown forms.By setting up local abnormalities,evaluating local abnormalities,and providing a new long-term memory method,the problem of local abnormality detection is defined.Finally,an ablation experiment was carried out to systematically improve the proposed network structure.Secondly,the copy-paste tampering data set is introduced to train Mask R-CNN,the Mask R-CNN network is improved to make it suitable for tamper detection,and the entire experiment process is optimized.Then,selected various tampered images as the data set,the types of objects tampered with in the data set include people,animals,furniture,buildings,etc.While satisfying the requirements of detecting target type diversity,the state of the target can be as diverse as possible.In terms of network structure,starting from the actual situation,according to the actual processor and data set,configured the network structure to obtain the best learning effect.Detected image copy movement to detect tampering area.Mask R-CNN has better performance,accuracy and less time consumption.The network model method has good scalability and can be extended to multiple format tamper detection technologies.Experimental results show that the method has good efficiency and accuracy.