Research On Pixel-level Image Tampering Detection Technology Based On Convolutional Networ

Image tampering detection technology is a hot research direction in the current field of information security.With the rapid development of image editing technology,people can use smartphones to manipulate images into false images that are difficult to discern with the naked eye,which not only brings convenience to people’s lives but also provides opportunities for criminals.Image tampering detection technology can effectively identify the authenticity and integrity of an image by analyzing the specific features of the image under test,thus providing effective security and stability for society.In recent years,numerous domestic and foreign experts and scholars have conducted extensive and in-depth research on this technology.Traditional tampering detection methods based on manually designed features require a large amount of prior knowledge,resulting in the model exhibiting good detection performance only on specific data sets.In contrast,image tampering detection technology based on deep learning utilizes the powerful representation learning ability of neural networks to automatically extract different types of tampering features without relying on heuristic experience design.Currently,image tampering detection methods based on deep learning mainly have the disadvantages of low detection accuracy and slow convergence.This article conducts in-depth research on the challenges of these image tampering detection methods and proposes corresponding solutions:(1)To address the problem of weak feature extraction capability in current copy-move tampering detection models,this article proposes a new image copy-move detection method.This method effectively extracts noise and edge information from the tested image through multi-angle feature fusion technology,and further improves the detection performance on image tampering edges by combining dilated convolutions and attention mechanisms.In addition,the model embeds tampering detection features into similarity features,enabling similarity detection to focus on specific areas,which effectively improves the detection efficiency and accuracy of model.Compared with existing copy-move detection methods,this method has strong robustness to various attacks while achieving good detection accuracy.(2)To address the problem of large structure and redundant parameters in existing tampering detection models using sliding windows,this article proposes a lightweight tampering detection method.This method replaces conventional convolutional modules with depth-wise separable convolutional modules on the basis of existing models to reduce model parameters,and uses the lightweight attention module CBAM to adaptively integrate local and global features,thereby minimizing model parameters while ensuring detection accuracy.At the same time,the model adopts a segmented guided training mode,significantly speeding up the network training convergence.Finally,a lightweight sliding window image tampering detection scheme with fast convergence and fitting is constructed.