Image Tampering Detection Based On Seam Carving

In recent years,with the rapid development of information technology,various digital image processing techniques have been widely used in people’s daily life,especially digital images,as one of the main carriers of information transmission,have become inseparable from our lives.Seam carving technology is a kind of content-aware image scaling technology,which has been successfully applied to digital image processing software such as Adobe Photoshop CS4 to resize images,and the images carved by this tampering technology are not visually obvious,which will make it difficult to judge the authenticity of the image content,so many unscrupulous people abuse this technology to tamper with images to satisfy their own selfish desires,such as deliberately removing objects from images to distort the true content of the images and distribute the tampered images arbitrarily.If the authenticity of these images is not confirmed,it is very likely to mislead the public,especially if the official media reports or the judiciary uses such images without knowledge,which will lose its credibility and notarization,and even cause social panic.Therefore,the detection of seam-carved images has been an important and attractive research topic for revealing the authenticity of digital image processing.Consequently,based on the characteristics of seam carving image tampering and the importance of its detection,two image tampering detection methods based on seam carved are proposed in this paper.(1)The image tampered with by the seam carving technique will leave some local traces,and based on the Squeeze-and-Excitation Networks(SENet)attention mechanism that can increase the captured property of feature map channel information,a convolutional neural network detection method based on SENet is proposed to determine whether the image has undergone seam carving.To obtain the ability to learn low-level features,a convolutional neural network architecture consisting of five types of network modules is designed to capture the subtle signals of the image,the SENet attention mechanism is introduced to increase the capture of channel information,and an integrated module is used to further enhance the performance and comprehensively analyze the features in a given local region for binary classification of the original image and the tampered image.The experimental results show that this algorithm achieves good detection accuracy.(2)Through the understanding of a large number of existing detection methods,we can know that current detection methods usually focus only on the local information of tampered images.To solve this problem,we propose a two-branch convolutional neural network seam carving image tampering detection method is proposed.One branch is a full convolutional network module for extracting local features of the image,and the Convolutional Block Attention Module(CBAM)is introduced to increase the capture of local information of the image;the other branch uses a fast Fourier convolutional model to complete the learning and extraction of global features of the image;Finally,the features extracted from the two branches are fused and input to the module used for binary classification.The experimental results show that this algorithm achieves good detection accuracy.