Image Inpainting Forensics Scheme Based On Deep Neural Network

With the continuous development and progress of society,digital image information has penetrated into every corner of social life,and the widespread use of digital images have also promoted the development and application of digital image editing software.Therefore,the operation of the digital image editing software becomes simple and convinent,so that the user can arbitrarily modify the image to achieve a better visual effect.However,the development of digital image editing software also allows some criminals to arbitrarily repair and tamper with digital image content,resulting in inpainting images in a large number of social life.Therefore,research on image inpainting forensics scheme based on deep neural network is of great significance for the development of digital images and the authenticity of image information.First of all,aiming at image inpainting forensic,a novel inpainting forensics scheme based on deep neural network is proposed in our paper.The scheme is composed of encoder network and decoder network,which does not limit the size of the input image,can predict the inpainting probability of each pixel in the image,and implements pixel-level forensics on the forensic image.At the same time,this paper uses the pyramid network to supplement the sampled output feature map information,and to refine the results of the forensic.At last,this paper analyzes the forensic performance of the different image inpainting method,compares the performance of different forensic schemes,and verifies the effectiveness of the scheme.Subsequently,to further enhance the generalization performance of the forensic scheme,this article analyzes the training data set of the forensic network and proposes to construct the training dataset by using the JPEG compressed image.Secondly,in order to improve the generation ability of the network and suppress the network over-fitting,this paper analyzes the regular term coefficient of the forensic network loss function.The correctness of the theoretical analysis was verified by experiments,and the detection results of forensics scheme were given,It shows that the forensic scheme has good detection performance and robustness.