Novac D (2012)
Publication Language: English
Publication Type: Thesis
Publication year: 2012
Publisher: FAU Erlangen-Nürnberg
One of the most common methods of manipulating an image is introducing new content. This typically requires adjusting of the newly inserted element to the content of the image by applying geometrical transformations such as resizing or rotating. These geometrical transformations are implemented as resampling operations, which make use of interpolation. This introduces artefacts in the image as correlations between neighbouring pixels, which makes it possible to detect this type of forgery. There are two main approaches in the existing literature, one based on the analysis of the residue of the derivative of the interpolated signal and the other based on the use of a linear predictor to determine the correlation coefficients between neighbouring pixels. We examine the advantages and disadvantages of these approaches by taking a closer look at the detectors proposed by Mahdian and Saic, based on the derivative approach and the one proposed by Kirchner, which uses the linear predictor method. We perform a quantitative and qualitative evaluation of these two methods under different tampering scenarios and different post-processing parameters. We extend our research by performing a differential analysis of the individual components of the detection pipeline and make an attempt to improve the detection in the particular case of lossy compression.
Novac, D. (2012). An Experimental Comparison of Resampling Detection Algorithms (Mid-study thesis).
Novac, Daniela. An Experimental Comparison of Resampling Detection Algorithms. Mid-study thesis, Erlangen: FAU Erlangen-Nürnberg, 2012.