Source Camera Identification Method Of Seam Carving Image Based On Indirect Correlation Coefficient

People are used to taking photos by digital cameras or smartphones in daily life.Various methods of picture editing and post-processing make it easier to edit and modify the photos,then spread to world through the Internet.In this regard,Judicial and public security organs also face huge challenges in criminal investigation and evidence collection.The method to find the source camera that took the tampered photos,so as to find the hands behind the tampered photos,has become a very important and practical research topic in the field of digital multimedia forensics.When the image is unprocessed,the traditional camera based on PRNU noise can be used to effectively identify the source camera.But after the image is edited and modified,the above method cannot identify the source camera directly and effectively.Seam carvi ng is an image scaling algorithm used to realize "content retention",it deletes or add seams selectively to expand or reduce images.The seam carving algorithm destroys the original camera fingerprint arrangement,so the source camera cannot be identified by the PRNU-based method directly.There are few existing studies on source camera identification of seam carving images.For this problem,this paper develops the following work:This paper proposes a source camera identification method for seam carving image based on indirect correlation coefficient.This method can find out which camera took the photos after seam carving among several known cameras.First,calculate the camera fingerprint of each camera according to a certain number of photos,and continuously increase the number of photos to obtain camera fingerprints under different photo numbers;second,for images after seam carving and have unknown camera sources,setting the energy threshold to obtain a sub-block from the image without seam carving,then,the sub-block is sent to the denoising filter.Subtract the original image and the output image of the denoising filter to obtain the noise residual,then,sliding the window continuously,the output image of the denoising filter is multiplied by the corresponding point of the camera PRNU,and then calculate the correlation coefficient between the product result and the sub-block noise residual.Finally,according to the correlation coefficient,select five features,use the entropy weight method to calculate the feature weights,then obtain the image source camera discrimination formula by weighting.This formula can express changes in multiple indicators related to the maximum correlation coefficient as number of images increases.When the experimental image and camera fingerprint come from different cameras,as the number of photos used to calculate the camera fingerprint continues to increase,the maximum value of the correlation coefficient does not have the above characteristics,and the coordin ates of the maximum value have no convergence trend,showing a random change characteristic.When the experimental image and camera fingerprint come from different cameras,as the number of photos used to calculate the camera fingerprint continues to incre ase,the maximum value of the correlation coefficient does not have the above characteristics,and the coordinates of the maximum value have no convergence trend,but change randomly.The decision formula of the source camera identification is given by cal culating the features and the weights of the features in this paper,accuracy of this source camera identification method reaches 88.75%.