Research On Forensic Techniques Of Image Provenance Analysis On Social Networks

Image provenance analysis is an important research direction in the field of digital forensics.Image source include historical source and device source.In order to explore the source of images,provenance forensics further divided into operation chain analysis and source identification.The goal of operation chain analysis is to identify whether the image is modified by a specific operation chain.The meaning of operation chain is a series of continuous operations,such as JPEG-crop-JPEG.The target of source identification is to identify the device source of an image,including camera,mobile phone,etc.Provenance forensics techniques have important applications in the judicial field,such as identifying the modification history and the device source of image to assist in the establishment of judicial evidence.At present,the operation chain is analyzed by statistical characteristics of the image,and the source identification methods find the source device of an image by taking sensor pattern noise as camera fingerprint.However,it is noted that the wide spread of images on social networks has decrease the performance of existing algorithms.Social networks have carried out unknown and complex processes on the exchanged image.These processes have seriously affected the statistical features and camera fingerprints that the existing algorithms rely on,resulting in the performance degradation of existing algorithms.Based on the above analysis,this paper carries out a series of researches on image provenance analysis techniques on social networks,mainly including the following three parts:(1)A new feature based on the matrix of block artifacts grayscale is proposed to analyze the crop operation chain of exchanged images on social networks.In view of the poor robustness of existing features after the image is exchanged through social networks,the changes of image block artifacts under aligned crop and non-aligned crop are analyzed,then the matrix of block artifacts grayscale is defined based on intra and inter block pixel correlation difference.The proposed new features significantly improve the performance of crop operation chain analysis in social networks and lossy recompression.(2)A new device source identification framework based on the reference pattern of social network(SRP)is proposed for exchanged images on social networks.This paper analyzes the change of image noise and camera fingerprint on social networks,and suggests using SRP as new camera fingerprints.SRP shows high robustness on social networks and is used to establish identification model,which improves the accuracy of device source identification on social networks.Meanwhile,the performance of tiny image source identification is significantly improved.(3)A new approach based on constrained convolutional network and and SRP is proposed to further improve the device source identificaiton accuracy of tiny images on social networks.By designing an adaptive constrained convolution network,the content-adaptive noise is extracted as part of device features.Then the content-adaptive noise is fused with SRP as the new device feature.The new features are used to train the device source classifier,which performed better than the state of arts in tiny image device source identification.