The Research Of Passive Forensics Against Various Image Retargeting Techniques

With the rapid development of computer technology,it has become easier to acquire and edit images.The advancement and innovation of image technologies can meet the needs of people for image editing,but also bring challenge to the authenticity and integrity of images.If image tampered is used on the Internet or scientific research,even as the court evidence,its harmfulness will be immeasurable.Therefore,image forensics is becoming one of the hot research areas.Image retargeting is a common type of image editing operation,it can also be used as image forgery for malicious purposes.Some image retargeting methods can alters the semantic content that an image conveys,or used in many complex image operations.Therefore,a universal detection approach is necessary to presented for the identification of image retargeting.Based on three representative retargeting techniques: seam carving,scaling and scale-and-stretch(scale-and-stretch),passive forensics algorithms are propose for the universal forensics and tampering history detection in this paper.The main contributions and innovations of the paper are as follows:Most existing detection approaches are designed for the specific image retargeting technique and treat the blind detection of retargeting as a binary classification.To solve the problem,a blind forensics approach is proposed for identifying various image retargeting techniques,not a single one.Since image retargeting operation usually leads to deformation on both local texture and global structure,which are different with the manipulation type,we proposed a 512-dimensional feature set,consisting of LBPCO(co-occurrence of adjacent local binary pattern)and BSIF(binarized statistical image features),to unveil the texture changes caused by image retargeting.Finally,support vector machine(SVM)classifier is exploited to identify the specific image retargeting technique.Compared with the existing methods,the detection accuracy of the proposed approach to detect single retargeting technique is greatly improved.And when the scaling ratio is between 10% and 50%,it can determine the type of different image retargeting with the detection accuracy of 92.97%~97.88%.For the detection of image retargeting and contrast enhancement operation chain,this paper inspired by steganalysis explores the statistics characteristics of tampering chain from the perspective of pixel correlation change.And it adopts the spatial rich model(SRM)with multiple residuals to carry out forensic work.Different tamper categories and sequences will have different effects on the inherent structure of the image,and the application of the rich residual sub-model in the algorithm can interpret the traces of pixel correlation from different levels.Finally,the support vector machine(SVM)is used as the classifier to detect and identify three different operation chains.The experimental results show that the algorithm can effectively identify the tampering category and tampering sequence on the seam carving-contrast enhancement operation chain,scaling-contrast enhancement operation chain and SNS-contrast enhancement operation chain with satisfactory detection accuracy.