Image Tampering Detection Based On Deep Learning

With the generalized use of high-resolution cyber-shot,users upload and share images every day.The spread of public images and the widespread use of image editing software have made the problem of image tampering increasingly serious.Image tampering is considered to be a potential threat,and the tampering person deliberately performs image tampering operations for insurance identification,judicial inspection,and network security.Therefore,it is very important to assure the credibility of the image.Digital image tampering detection technology is different from other image detection.It needs to pay more attention to tampering features.Therefore,the key to research is whether it can extract image tampering features that meet the purpose of forensics.Traditional hand-designed features often restrict the generalization ability of the model.In recent years,as deep learning is applied to more and more scenarios for image forensics,deep neural networks have been able to be used for image forensics.When creating a tampered image,the tamper can use different image editing operations to tamper with the image,and the inspector must test each editing operation.Therefore,people are more interested in developing a general algorithm that can detect multiple tampering methods.In order to resolve these issues and achieve a more general detection model,this paper improves some of the current advanced network models,and proposes image forensic algorithms and image forensic positioning algorithms.First of all,the extraction of features in tampered images is a difficult point in the task of image forensics,and how to design an effective network that can extract tampered features more abundantly has become the focus.Therefore,in view of the problem of image tampering forensics,this essay proposes a multi-scale CMDN network with constrained convolutional layer by improving the Dense Net network.Experiments show that the model achieves 97.90% accuracy on the data set VTDS-2007 data set under the condition that the accuracy is the evaluation standard,which proves the feasibility and effectiveness of the method.In addition,the positioning problem for image forensics is different from semantic object detection.It needs to pay more attention to the distinguishing features between tampered areas and non-tampered areas,which indicates that the network needs to learn richer features.Therefore,we propose a two-branch Mask R-CNN network with Attention.One of the branches is the main branch.The purpose is to use the attention mechanism to extract features from RGB images to find traces of tampering,such as strong distinction differences and abnormal tampering boundaries.The other is the noise branch,which uses the noise features extracted by the Steganographic Enrichment Model(SRM)filter layer to distinguish the noise inconsistency between the real area and the tampered area.Finally,the bilinear pooling layer(Bilinear Pooling)combines the characteristics of the main branch and the noise branch,and further learns the information on the two branch spaces.The algorithm has been tested on three public data sets,and the detection performance is better than the unimproved Mask R-CNN network.At the same time,the algorithm is a general model structure that realizes classification,location,and segmentation of tampered areas.