Image Splicing Tampering Detection Based On Double Branching Feature Fusion

With the popularity of smart imaging devices,people are able to modify and deliver digital images at low cost.When forged images are widespread,they will have negative effects,therefore,image forensics technology is also given more and more attention.Image splicing tampering is one of the most common ways of tampering.By moving parts of the original image directly or indirectly to the tampering image to achieve the purpose of tampering,and hiding the tampering traces after processing.Tamper detection usually includes three main steps: image pre-processing,feature extraction,feature classification and localization.The existing deep learning-based graph splicing tampering detection algorithm can extract the discriminant features of images,but there are still some problems such as insufficient tampering feature extraction,low detection accuracy,and inaccurate tampering image positioning with small splicing area.Based on this,this paper presents a study on image splicing tamper detection based on double branching feature fusion.It specifically includes the following two parts:(1)In view of the problems of insufficient image tampering feature extraction and inaccuracy of splicing location region,a two-branch image splicing tampering detection model with fusion depth clustering is proposed.The model first uses the tamper image and its corresponding high-pass filtered image as network input,and uses the feature pyramid networks in the image feature extraction stage.Secondly,a deep clustering model is introduced to connect with the backbone network,and the extracted feature map is optimized by clustering to retain the tamper regions of the image and remove the non-tamper regions of the image.Finally,the network is recommended to locate the tamper area through the region.Comparing the public data set with other models,the results show that the evaluation index of this model is higher than that of other models,which can effectively detect the splicing tampering area of the image,and is relatively stable against tampering JPEG compression,rotation and zoom attacks.(2)In view of the problem of ignoring the tampering image with small splicing area,a double-branch image splicing tampering detection model based on multi-layer feature refinement fusion is proposed.The model first uses the tampered image and its corresponding filter-processed image as input to the double-branch network,and refines the fusion network through multi-layer features to integrate the deep refinement features of the image with the shallow refinement features,so as to improve the accuracy of splicing tampering detection in small areas.Finally,for the extracted double branching features are fused through a bilinear pooling layer.On this basis,the anchor box prediction structure and adaptively adjust the different shapes of the anchor box to further improve the accuracy of model detection.The experimental results show that the proposed method is more accurate than several existing methods,and can effectively deal with JPEG compression,rotation,scaling,blur and other attacks.