Detectin Double JPEG Compression Based On Deep Learning

With the increasing popularity of digital cameras,smart phones and other imaging devices,image has become an important information carrier in today's society,and plays an important role.With the growing power of image processing software,users can easily change the image content without leaving visual traces.With the development of Internet,these tampered images may account for tremendous social impact.Therefore,more and more scholars pay attention to the passive forensics technology.JPEG is the most widely used image compression format.Since image tampering is usually accompanied by double JPEG compressions,detecting the history of JPEG compression is an important research branch.Detection of double JPEG compression with same quality factor has been regarded as a challenging task in digital image forensics because there are very few modification cues in the tampered images,especially when the compression quality factor is low.Convolutional Neural Networks(CNN)has achieved excellent performance beyond traditional methods in many areas of computer vision.In this paper,CNN is applied to double JPEG compression detection with the same quality factor.And a network based on dense connection is proposed.With the appropriate network design and contributing to the characteristics of average pooling,dense connection and transition,the proposed network,which is not related to the image content,can differentiate double JPEG compression artifacts accurately.Further,through the analysis of residual connections and dense connections,this paper proposes a network that combines these two connections.Each dense connection block is grouped with a residual connection block.The network consists of three groupings.This structure can further reduce the complexity of the network while retaining the image features extracted from the shallow layer.Finally,the CNN feature and traditional feature are combined by cascade method,and the combined feature is trained and tested based on machine learning framework to further improve the accuracy of the algorithm.Experiment results on the two datasets have demonstrated that the proposed feature outperforms several state-of-the-art approaches investigated.In addition,the second network structure also achieves excellent results in adaptive JPEG image steganography detection.