The Research For Digital Image Resampling Detection

As the sophisticated digital image processing software are widely used in our daily life, they have allowed the production of image forgeries that are virtually indistinguishable from authentic photographs. The research of digital image forensics, as the forefront of the emerging international research field, is therefore of practical interest and value.The creation of convincing image forgery oftentimes involves geometric transformations, which typically comprise a resampling of the original image to a new sampling grid. Therefore, amongst others, resampling detection has become a standard tool for the forensic analysis of digital images. In this paper, we focus on digital image resampling detection techniques for digital photographs rendered by digital cameras.We start from analyzing the the resampling algorithms and the statistical principles for resampled images, and then two classical algorithms are introduced thoroughly. After programming and testing, we find some defects and limitations. Last, we propose two new detecting algorithms for the defects and limitations of classical detecting algorithms.The main contribution of this paper is to propose two new algorithms. The first method for detecting image resampling is proposed based on the distribution of the most significant digits of the AC coefficients in DCT domain. By calculating the above distribution, we use weighed absolute value as a feature to indicate the marked difference after resampling. Experimental investigations demonstrate that the proposed approach is able to detect resampling efficiently.In the second method, we notice that when we upload an image to a website, the uploading process include a resampling operation followed by a recompression operation. And this process will inevitably disturb the statistical features of the original image. So we propose a method based on statistical features to detect whether an image has been uploaded. First, we establish a model based on statistical features extracted from the given test image. Then Support Vector Machine ( SVM ) is chosen as a classifier to train and test the given images. Experimental results demonstrate that this new scheme can detect the uploading process effectively.At the end of the thesis, the prospect of blind digital image forensics is discussed subsequently as well as the possible future research areas.