Research On Image Copy-Move Forgery Detection Based On Deep Learning

Along with the development of the times and the advancement of technology,various digital photo equipments are continuously updated and the functions of various image editing tools are becoming more and more powerful.Although the operation of picture editing software is simple and convenient,at the same time,maliciousing and picture information forgery have become a very serious and thorny problem in modern information systems and daily life.This phenomenon has widely penetrated into important scenes such as academic research,criminal investigation,military field,news report and even national level,and has brought extremely serious adverse consequences to these fields.Therefore,it is very important to protect the authenticity of digital image media and detect and locate the image forgery.In recent years,deep learning techniques have become an area of high-profile research.Deep learning is also widely used in image detection and has achieved good results.Image copy-move is a very common and difficult to detecting attack method in image.This paper proposes a method based on deep learning techniques for detecting copy-move forgery.The main work is as follows:1.Aiming at the classification of whether there is copy-move forgery in the image and the location of the similar region of the fake image,a multi-scale convolutional neural network based on feature pyramid network(PFN)is proposed.This network can not only realize the binary classification of the authenticity of the input image to detect the fake image,but also perform pixel-level positioning on the original area and similar areas in the fake image,generate a Ground Truth mask with the same size as the original input image,and detect Forgery regions in the image.The proposed network improves the problems of single field of view and poor robustness existing in existing methods.The accuracy of binary classification detection of images is high.On different data sets,for image forgery areas of different sizes,it can deal with attacks such as noise Gaussian and geometric scaling.Compared with other literature algorithms,the performance indicators still maintain a high performance.It can be better applied in actual scenarios.2.For images with copy and move forgery,the copy source area and move target area positioning problem.A localization network based on EfficientNet convolutional neural network is proposed.It can efficiently classify similar regions in the ed image,and locate the copy source region and move target region in the fake image.EfficientNet scales the depth,width and resolution of the network in a balanced manner.Compared with other deep learning algorithms,it has higher performance in terms of detection accuracy and robustness.The similar area localization network and the image source area and target area localization network are connected in series to realize accurate localization of ed source and target area in the fake image,which has higher accuracy and time efficiency than parallel networks in other literatures.