The Research Of Passive Forensics Against Image Seam Carving

In recent years,digital images have been playing an important role in our daily work and life.With the popularity of image editing software,ordinary users can easily produce faked images without leaving any annoying artifacts.If these faked images are used in the news media,scientific discovery,court evidence and other important fields,it will take tremendous harm to society.So it is so important to design a forensics algorithm to detect the authenticity and integrality of the image effectively.As a popular content-aware image retargeting technique,seam carving has been implemented in many image editing software,such as Meitu,Adobe Photoshop and GIMP.Through these image editing software,some criminals can easily use seam carving technology for image forgeries such as object removal.Since seam carving can be used to change the semantic content of the original image without leaving obvious tamper clues,which leaves a huge challenge to our forensics task.In this paper,we focus on the passive image forensics for seam carving,and the main work and contributions are as follows:Firstly,this paper puts forward an improved forensic method to detect seam carving based on the three-element joint density of the difference between adjacent pixels.It is clearly that there is a strong correlation between adjacent pixels in the original image.But seam carving inevitably changes the neighboring relations of pixels near a deleted seam,which will influence the distribution of the difference between adjacent pixels?Finally,we are motivated to use three-element joint density to model the differences between adjacent pixels as forensics features,and the Support Vector Machine(SVM)is used as a classifier to determine whether an image is original or suffered by seam carving.The experimental results show that our proposed method improves the average detection accuracy,and the method is robust against seam carving with JPEG post-compression processing.Secondly,there is still some scope to further improve the detection accuracy and stability to detect seam carving with low scaling ratios for existing research.This paper presents a new forgery detection method based on block reshuffle a nalysis.When a seam is carved for image shrinkage,there are pixel displacements to fill in the gap left by this seam.Specifically,each pixel at the right side of or below this seam is reshuffled into non-overlapped 8×8 block,which changes the distribution of local image entropy.Instead of directly extracting local image entropy-based features from candidate image,image cropping is firstly conducted as pre-processing.Due to the self-alignment mechanism of image cropping,it is possible to makes an alignment for reshuffle blocks with no seam deleted in the carved image.This greatly alleviates the side effect of image content on detection accuracy.The experimental results prove that our proposed method achieves better detection performance for seam ca rving with low scaling ratios.