The Research Of Passive Forensics Against Image Seam-carving

With the proliferation of image editing software,novices can easily tamper with image content and leaves no visible traces.Considering the significance of image authenticity,images are often used as news,scientific research results,medical applications and judicial evidence,which has been mixed with tampered images,then the consequences are hard to imagine.Therefore,manipulation detection that confirms the authenticity of the image is necessary.Seam carving can effectively scale the image,and the scaled image will not have visible clues.Currently,people can use this technology through Adobe Photoshop.At the same time,it is also used by some criminals for image tampering,such as the deletion of specific objects in the image.The existing algorithms can solve the manipulation detection of various scaling ratios,but there are still problems in the low scaling factor performance and robustness.In order to solve the above problems,this paper proposes two detection algorithms based on joint features.Aiming at the seam carving detection,a joint feature algorithm based on local binary pattern(LBP)energy bias features in the residual energy domain and Markov features is proposed.The main idea behind that is a certain correlation between adjacent pixels in the image.In the seam carving operation,each time a seam is removed from the image,the pixels adjacent to the right side of the pixel must be shifted to the left to fill the removed pixel position.Then the relationship between adjacent pixels of the image removal seam changes,and at the same time,the value relationship between the pixels may be influenced.Finally,the extracted joint features are input into a support vector machine(SVM)classifier,and the extracted features are trained and classified.Experiment shows that the algorithm achieves better detection performance regardless of low scaling factor or high scaling factor,and can resist to post-processing operations such as JPEG compression.Most of the existing detection algorithms have problems with unstable detection performance.This chapter uses high-order probability statistical distribution to further describe the relationship between pixels,that is,the difference pixel adjacency matrix(SPAM)feature.The SPAM feature can reflect the traces left by the JPEG pre-processing operation in the JPEG-seam carving.Combined with the energy-based and patch-based statistical distribution features proposed in Chapter 3,the differences in the image after tampering are described.Experiments show that the algorithm caneffectively distinguish the original image,the seam carving tampered image and the seam carving image that has undergone pre-processing operation JPEG compression.In addition,the stability of the algorithm is improved and the detection accuracy decreases slightly when the scaling is small.